【双语音+文稿】（完整版）2026 英伟达GTC柏林大会：英

共 2 集

剧集目录

集	标题	时长
P1	【双语音+文稿】（完整版）2026 英伟达GTC柏林大会：英	1:33:26
P3	【双语音+文稿】（完整版）2026 英伟达GTC柏林大会：英	1:33:26

P1 · 【双语音+文稿】（完整版）2026 英伟达GTC柏林大会：英伟达创始人兼CEO黄仁勋发表主旨演讲！ p01 英文版 (P1)

【双语音+文稿】（完整版）2026 英伟达GTC柏林大会：英伟达创始人兼CEO黄仁勋发表主旨演讲！ p01 英文版 (P1)

发布于 2026-10-22 · 时长 1:33:26

字幕摘录

时间	英文	中文
0:09	This is how intelligence is made a new kind of factory	这就是智能制造的新型工厂
0:16	generator of tokens	符号生成器
0:19	building blocks of AI	A. 大赦国际的组成部分
0:25	Tokens have opened a new frontier	托肯斯开辟了新的疆域
0:29	Turning data into knowledge and knowledge into action	将数据转化为知识转化为行动
0:38	They decipher nature's code to reveal hidden cures	他们破译自然的密码揭示隐藏的解药
0:48	And mirror our cities	照亮我们的城市
0:57	And tirelessly navigate our world	不知疲倦地穿越我们的世界
1:01	To keep us moving	继续前进
1:06	Tokens learn from the way we move	托肯斯从我们的行动中学习

展开字幕全文（1232 条）

序号	英文	中文
1	This is how intelligence is made a new kind of factory	这就是智能制造的新型工厂
2	generator of tokens	符号生成器
3	building blocks of AI	A. 大赦国际的组成部分
4	Tokens have opened a new frontier	托肯斯开辟了新的疆域
5	Turning data into knowledge and knowledge into action	将数据转化为知识转化为行动
6	They decipher nature's code to reveal hidden cures	他们破译自然的密码揭示隐藏的解药
7	And mirror our cities	照亮我们的城市
8	And tirelessly navigate our world	不知疲倦地穿越我们的世界
9	To keep us moving	继续前进
10	Tokens learn from the way we move	托肯斯从我们的行动中学习
11	Deliver the care we need	提供我们需要的护理
12	And walk into danger	走进危险之中
13	To protect the ones we love	为了保护我们爱的人
14	They see what we cannot	他们看到我们所不能
15	So the steadiest hands never work alone	所以最稳健的手永远不能单独工作
16	Every idea Tokens spark	每一个想法托肯斯火花
17	Every success they share	他们分享的每一个成功
18	Puts what matters within reach	将重要的事物放在眼前
19	Tokens are helping us break new ground	托肯斯帮助我们开拓新天地
20	On a scale never seen before	在前所未有的规模上
21	So we can reach	这样我们就能到达
22	For the sky and beyond	为了天空和天空之外
23	Together we take the next great leap	我们一起走下一个大跃进
24	Into a bright new future	进入光明的新未来
25	Built for all mankind	为全人类而建
26	And here is where it all begins	在这里,一切都开始
27	欢迎来到舞台 NVIDIA创始人兼CEOJensen Wong.
28	Welcome to GTC.	欢迎来到GTC.
29	It's great to be here.	赞曰.
30	Before we start, let's give it up for our pre-game show hosts, smart and adorable, smart and adorable.	在开始前,让我们放弃我们的预赛节目主持人,聪明可爱,聪明可爱.
31	They spoke about AI infrastructure, sovereign AI, AI safety, and physical AI.	他们谈到了AI基础设施,主权AI,AI安全,以及实体AI.
32	They covered an enormous amount of ground this morning, and we are going to take it one step further.	今天上午,它们覆盖了大量地面,我们将进一步推进。
33	有趣的是,L'Oreal、Revolut、Spotify、Merck、Vodafone、Zara、保时捷、Nestle和汇丰银行有什么共同点?
34	Well, they're all European companies and they're all working with NVIDIA on AI.	他们都是欧洲公司他们都和NVIDIA合作研究AI
35	Everybody, everybody here is all activated on the AI ecosystem and working towards the AI economy.	各位,这里的每个人都在人工智能生态系统上启动并致力于人工智能经济
36	We've been working here in Europe now for 33 years.	我们已经在欧洲工作了33年
37	You probably don't realize that NV1, NVIDIA's first chip, NV2, and NV3, which became Revot 128,	你可能没有意识到NV1,NVIDIA的第一个芯片,NV2和NV3,它变成了Revot 128,
38	the chip that saved our company really, and launched our company, were all built in Europe.	真正拯救我们公司的芯片, 并推出我们的公司, 都是在欧洲建造的。
39	我们的第一个搭档是SGS Thompson
40	We manufactured a chip in Kroll, France, assembled it in Malta, and shipped it around the world.	我们在法国的克罗尔制造了芯片, 在马耳他组装,并运至世界各地。
41	Our first largest customer, the largest customer of GeForce Days in the beginning, was a company called MediaMark.	我们的第一个最大的客户,GeForce Days在一开始最大的客户是一家名为MediaMark的公司.
42	Do you know them?	你认识他们吗?
43	MediaMark had a spectacular idea about building PCs, as you remember, and they were our largest customer.	MediaMark对建造个人电脑有一个惊人的想法, 你还记得, 他们是我们最大的客户。
44	It was Elsa, a German company that brought NVIDIA into the workstation industry in the year 2000.	是德国公司Elsa在2000年将NVIDIA带入工作站产业.
45	Now, NVIDIA is in every single workstation all over Europe designing cars,	现在,NVIDIA在每一个工作站在整个欧洲设计汽车,
46	everything from styling, design, engineering, simulation, all the way through to even inside the car,	从造型、设计、工程、模拟到车内
47	which a European company brought us into, Audi, in the year 2005.	是一家欧洲公司在2005年带我们来的奥迪
48	Then in year 2020, we started working on autonomous vehicles with Mercedes.	之后在2020年,我们开始与梅赛德斯一起从事自主车辆的工作.
49	As you can see, NVIDIA's life journey, the entire 33 years, I've spent a lot of time in Germany,	如你所见,NVIDIA的人生历程, 整个33年,我花了很多时间在德国,
50	so I'm delighted to be back.	所以我很高兴能回来
51	Thank you.	谢谢
52	I'm delighted to do GTC here.	我很高兴在这里做GTC。
53	This year, this last year, we started the new journey to building a new industry here in Europe,	今年,去年,我们开始了在欧洲建设新产业的新征程,
54	the beginnings of the AI economy.	AI经济的开始.
55	In the last year alone, we've invested nearly $10 billion in Europe across startups,	仅在去年我们在欧洲投资了将近100亿美元跨创业企业
56	big partnerships, small partnerships, all over Europe.	大型伙伴关系,小型伙伴关系,遍布欧洲。
57	This last week alone, as I journeyed from France and UK to Sweden and now here in Berlin,	仅上周,当我从法国和英国前往瑞典,现在在柏林,
58	we announced the investments of 14 startup companies.	我们宣布了14家创业公司的投资
59	Overall, since the beginning, we now have 12,000-plus partnerships across Europe,	总体而言,自一开始,我们在欧洲各地建立了12,000多个伙伴关系。
60	and we're all working towards one goal now.	我们都在努力争取一个目标
61	After decades, after decades of advancement in the computer industry,	经过几十年的计算机工业的进步,
62	we are now at the beginnings of a new industrial revolution.	我们现在正处于新的工业革命的开始阶段。
63	This image was generated by a computer.	这个图像是由计算机生成的.
64	It was prompted by me.	是因为我
65	Nice, simple prompt about AI powering a new industrial revolution generated this.	人工智能发动新的工业革命的简单而明智的提示导致了这个结果
66	I gave it a few partners that I wanted to highlight, companies that we're working with,	我给了它几个合伙人我想强调, 我们合作的公司,
67	and this is an extraordinary thing.	这是一件不寻常的事。
68	In the mind of the AI, this is what it means to build a new industry,	在AI的心目中,这就是建设新产业的意义,
69	generation of tokens, the production of intelligence,	创造象征创造智慧
70	which integrates into every single industry powering a new industrial revolution.	它融入每一个工业推动新的工业革命。
71	A new industrial revolution happens when the new technology affects how we build something	新的工业革命会发生当新技术影响我们如何建造东西
72	and what we can build, how we build something and what we can build.	以及我们能够建造的东西,我们如何建造东西和我们所能建造的东西。
73	I think in every single way, artificial intelligence will affect everything we build	我认为,在每一种方式, 人工智能会影响我们建立的一切
74	and in fact enable us to build new things.	并且事实上使我们能够创造新的东西。
75	In this way, because everyone will use it, every company will be powered by it,	这样,因为每个人都会使用它, 每个公司都会得到它的力量,
76	because it's digital intelligence, the foundation of everything.	因为它是数字智能, 一切的基础。
77	Every single person will use it, every company will be powered by it,	每个人都会用它每一个公司都会用它
78	every industry will be transformed, and new ones will be created.	每一个产业都会被改造,新的产业也会被创造出来.
79	In this way, like the electricity, it is a new industrial revolution,	这样,就像电力一样,这是一场新的工业革命,
80	and we're at the very beginnings of that.	我们正处在一个开始。
81	The question is, what is AI?	问题是,AI是什么?
82	And this is a big deal.	这是件大事。
83	If you want to be part of creating this future, understanding what is AI is vitally important.	如果你想参与创造这个未来,了解AI是什么至关重要。
84	And this is really a simple framework.	这是一个很简单的框架。
85	We were early to observe that in fact AI is a manufacturing industry.	我们很早就注意到,事实上AI是一个制造业。
86	AI is not just software technology.	AI不仅仅是软件技术.
87	It is software technology, but not just.	这是软件技术,但不只是。
88	AI, of course, runs on a computer, but the way that that computer operates is fundamentally different.	当然,AI运行在一台计算机上,但该计算机的操作方式根本不同.
89	Rather than a tool that's used by us, it is now part of a machinery and infrastructure,	现在它不是我们使用的工具而是机器和基础设施的一部分
90	factories that are producing intelligence that are integrated into every single industry and application.	生产情报的工厂已经融入每一个行业和应用
91	A simple model is to think of AI as a five-layer cake.	一个简单的模型是把AI看作是一个五层的蛋糕.
92	Energy powers advanced computers that are part of an AI factory that generates tokens,	能源动力高级计算机属于AI工厂的一部分生成符号,
93	digital intelligence from the models that run on it that are integrated into every application.	来自运行在它上的模型中的数字智能被整合到每一个应用程序中.
94	A five-layer cake.	五层的蛋糕
95	To be part of this AI economy, you don't want to just be consuming the AI.	为了参与AI经济,你并不想仅仅消耗AI.
96	You want to be producing the AI.	你想制作人工智能。
97	We could do all of that here in Europe.	我们可以在欧洲做这一切。
98	And in fact, we've already started.	事实上,我们已经开始了。
99	Europe is building its AI economy and its five-layer cake represented here.	欧洲正在建设其AI经济及其代表在这里的五层蛋糕.
100	We've got great partnerships, and when you look across this five-layer cake,	我们有很好的合作关系, 当你看到这个五层蛋糕,
101	you can recognize that in fact it is a new industry because there's growth at every single layer.	你可以认识到,事实上这是一个新的产业, 因为每一个层次都有增长。
102	We're creating jobs, creating growth at an extraordinary rate.	我们正在创造工作机会, 以非凡的速度创造增长。
103	Companies here in Europe, Schneider, Siemens Electric, Siemens Energy, doing incredibly well around the world.	欧洲的公司,施耐德公司,西门子电气公司,西门子能源公司, 在世界各地都表现非常好。
104	We're the computing layer.	我们是计算层。
105	We integrate into AI factories that are now flourishing all over Europe.	我们融入了整个欧洲的人工智能工厂。
106	Startup companies like Nebia is doing incredibly well.	像Nebia这样的创业公司表现的非常好
107	A new company called Firebird, which is setting up a new cloud in the Central Asia region.	一家名为火鸟的新公司,正在中亚地区建立新云.
108	And even Cassava in Africa.	甚至非洲的木薯。
109	Every single telco in the future, and I predict this, every single telco in the future,	未来每一个Telco, 我预测,未来每一个Telco,
110	with a hundred thousand sites representing some hundred gigawatts of distributed telco base stations and data centers,	拥有10万个站点,代表着大约100千兆瓦的分布式电视机基站和数据中心,
111	will in the future evolve into a distributed AI network.	未来将发展成分布式AI网络.
112	So the AI infrastructure layer is being built out, which is then above that, the AI models.	因此AI基础设施层正在建设中,然后就超过了这个,AI模型.
113	It's really important to recognize that long before the AI models emerge, the infrastructure has to exist.	真正重要的是要认识到,早在AI模型出现之前,基础设施就必须存在.
114	Just as before you can have an internet, you have to have the networking.	和之前的互联网一样, 你必须有网络。
115	Just as before you could really enjoy the capabilities of electricity, you have to have electrified houses, electrified power plants.	就像以前你真正能够享受电力能力一样, 你必须拥有电气化的房屋,电气化的发电厂。
116	And so the models are the next layer.	所以模型是下一层。
117	It's also interesting to observe that the model layer is not just language models.	同样有趣的是观察到模型层不仅仅是语言模型.
118	It represents language, biology, chemicals, physics, robotic actions, animation, and it represents models of all sizes.	它代表语言,生物学,化学,物理,机器人动作,动画,它代表各种大小的模型.
119	These different industries are going to be powered by different modalities of AI models,	这些不同的产业将会通过不同模式的AI模式获得动力,
120	which then integrate into the AI natives, the startups.	然后融入人工智能本土,即初创企业。
121	And there are so many here in Europe.	欧洲有很多人
122	So the Europe's AI economy is starting, and it's really accelerating.	因此,欧洲的AI经济开始,它真的在加速.
123	It's great to see.	赞曰.
124	At the foundation of it, where we contribute, most people think is GPUs.	在它的基础,我们在那里贡献,大多数人认为是GPU.
125	It is true.	这是真的。
126	We started in 1993 building a graphics chip, which emerged into the GPU, the programmable shading GPU.	我们从1993年开始建造图形芯片,它出现在GPU,可编程阴影GPU中.
127	Over time, we added CUDA acceleration to that, reinvented computing as we know it, which then brought forth artificial intelligence.	随着时间的推移,我们添加了CUDA加速度, 重新发明了我们所知的计算, 从而带来了人工智能。
128	But NVIDIA today is really a full stack, end-to-end, complete AI factory platform.	但今天的NVIDIA其实是一个完整的堆栈,端到端,完整的AI工厂平台.
129	The breadth of the technology of our company over the years has really expanded.	这些年来,我们公司技术的广度确实得到了扩大。
130	Everything from GPU computing, Vera Rubin is our latest generation, to the NVLink switches that connect them into one giant GPU.	从GPU计算,Vera Rubin是我们最新的一代, 到连接到一个巨型GPU的NVLink开关。
131	To the Vera Bluefield 4 compute that we recently introduced, and I'll say a few words about that later.	Vera Bluefield 4 计算一下我们最近介绍的, 我稍后会说几句话。
132	To Spectrum X, which is our Ethernet solution, making NVIDIA the largest networking company in the world today.	致Spectrum X,即我们的以太网解决方案,使得NVIDIA成为当今世界最大的网络公司.
133	To Grok 3, a decode accelerator for token generation.	给Grok 3,一个解码加速器,用于代号生成.
134	Vera Bluefield 4的存储加速, 我马上会说几句话。
135	And then Vera RTX compute, and I'll mention that in just a second.	然后Vera RTX计算, 我马上会提到这一点。
136	But this now represents the breadth of technology necessary to create the AI factory.	但这代表了创建AI工厂所需的技术的广度.
137	But nothing works unless you have software that activates it all.	但是,除非有软件激活它,否则一切无效。
138	The first part is the system software that runs the AI factory.	第一部分是运行AI工厂的系统软件.
139	These systems are gigantic.	这些系统是巨大的。
140	A 1 gigawatt data center, 1 gigawatt AI factory, costs about $50 billion.	1千兆瓦数据中心,1千兆瓦AI工厂,耗资约500亿美元.
141	Land, power, shell, all of the compute, networking, storage, power, cooling, all connected together, multiple acres of AI factory.	土地,动力,壳体,所有计算,网络,存储,动力,冷却,全部连接在一起,多亩的AI工厂.
142	Just turning it on and operating it is a marvel in itself.	打开和操作它本身就是一个奇迹。
143	The magnificent amount of software necessary to energize this factory, to put it to use and produce tokens is an extraordinary feat.	为了给工厂注入活力, 使其投入使用和生产代币, 所必须的软件量是非凡的功绩。
144	NVIDIA offers the entire full stack software.	NVIDIA提供完整的堆栈软件.
145	On top of that are what we call the CUDAX libraries, the algorithms of our company.	除此之外,我们称之为CUDAX库, 我们公司的算法。
146	What NVIDIA truly is, is an algorithm company that designs silicon to accelerate them.	NVIDIA真正的是, 一个算法公司设计硅加速它们。
147	Of course, one of them is cuDNN, what many researchers recognize as the library that brought forth, that started the big bang of modern AI, cuDNN.	当然,其中一个是CuDNN, 许多研究者都认为它是一个图书馆, 它带来了现代AI,CuDNN的大爆炸。
148	It's cuDNN is to AI what SQL is to the world's database industry.	它的cuDNN是AI什么SQL是世界数据库产业.
149	It's a domain-specific language, a domain-specific library.	它是一个域特有语言,域特有库.
150	Just as OpenGL and DirectX enabled real-time computer graphics, cuDNN enabled the AI computing industry.	正如OpenGL和DirectX启用实时计算机图形一样,cuDNN启用了AI计算行业.
151	350 other libraries, those libraries are in fact the crown jewels of our company.	350个其他图书馆,这些图书馆实际上是我们公司的皇冠珠宝。
152	Everything from computational lithography to signal processing, image processing, data processing, particles, fluids, sparse solvers.	从计算文学到信号处理图像处理数据处理粒子流体零散的解析器
153	Incredible breadth of libraries that many of you in the audience have connected into your company.	图书馆的宽度令人难以置信你们中许多人都与你的公司有联系
154	This is how we open new markets.	这就是我们打开新市场的方式。
155	This is how we connect NVIDIA into algorithms that enable you to do computation at incredible speeds.	这就是我们如何将NVIDIA连接到算法中,使你能够以不可思议的速度进行计算.
156	And then finally on the right-hand side, on that side, the models.	最后在右侧, 在那边,模型。
157	This is the new software, the new engine of new applications.	这是新软件,新应用程序的新引擎.
158	And it sits on top of and it collaborates with the CUDAX libraries.	它坐落在顶端,并与CUDAX图书馆合作.
159	These models NVIDIA have been working on across multiple domains.	这些模型NVIDIA一直在跨越多个领域开展工作。
160	With these models, it opens up new opportunities.	通过这些模型,它开辟了新的机遇.
161	And I have lots to say about that later on.	我以后有很多话要说
162	每一代人 NVIDIA都是从Pascal开始的
163	The AI era started with the Pascal new, what is called DGX1 system.	AI时代始于帕斯卡尔新式,即所谓的DGX1系统.
164	DGX1 brought to the world the world's first AI supercomputer.	DGX1为世界带来了世界上第一台AI超级计算机.
165	I introduced it 10 years ago, delivered the first one to OpenAI, started their computation journey.	我10年前介绍了它, 给OpenAI送了第一个,开始他们的计算旅程。
166	From Pascal to Volta to Ampere, then to many of you who heard about us for the very first time.	从帕斯卡到沃尔塔到安佩尔,然后向你们中许多第一次听到我们的消息的人.
167	Hopper, the current generation Blackwell, and then now Vera Rubin in full production.	霍伯(Hopper),即现代布莱克韦尔,后来薇拉·鲁宾(Vera Rubin)开始全面生产.
168	Notice six generations of architectures.	注意到六代建筑.
169	With each generation, we advanced the technology.	每一代人,我们都会进步技术。
170	In the course of that last 10 years, we've improved AI performance by one million times.	在过去十年中,我们把AI的性能提高了一百万倍.
171	If you would take a step back and think about what would Moore's law have done,	如果你退一步想想摩尔的法律会做什么
172	in 10 years' time, the first five years would have improved by a factor of 10,	10年后,头5年会改善10倍
173	the second five years by another factor of 10, 100x improvement in 10 years' time.	后五年又增加了10倍 10年后100倍
174	In the same period of time, because of the way NVIDIA Extreme Codesigns	在同一期间,由于NVIDIA极端代号的方式
175	were innovating across every single computer, across every single stack of the software,	每一个电脑,每一个软件堆,
176	optimizing new algorithms, creating new algorithms, while creating architectures to accelerate it,	优化新的算法,创建新的算法,同时创建架构来加速它,
177	working across the entire domain, what we call Extreme Codesign,	在整个领域工作, 我们称之为极端代号,
178	we were able to accelerate AI by a million times, and we're going to continue to do so.	我们把AI加速了100万倍,我们将继续这样做。
179	This is what NVIDIA builds today, and this is the foundation by which we can enable Europe	这就是NVIDIA今天所建立的基础,这是我们能够使欧洲得以发展的基础。
180	to be part of the AI ecosystem, the AI economy.	成为AI生态系统,AI经济的一部分.
181	Now, each generation, in the beginning, Pascal was GPUs, Volta was GPUs, Ampere was GPUs,	现在,每一代人,在一开始,帕斯卡尔是GPU,沃尔塔是GPU,安佩尔是GPU,
182	and Hopper was GPUs.	霍普是GPU的员工
183	We invented the new DGX form factor.	我们发明了新的DGX形式因子.
184	Then it was Blackwell that introduced to the world the first rack-scale computing,	之后是布莱克威尔向世界介绍第一个架式计算,
185	and I'll show you that in just a moment.	我马上给你看
186	Rack-scale computing turned one entire system into one giant computer, one GPU,	拉克规模的计算把整个系统变成了一个巨大的计算机,一个GPU,
187	and now, with Vera Rubin, we're extending it to processing beyond models,	而现在,与Vera Rubin, 我们正在扩展它处理超出模型,
188	beyond large language models, to now processing the entire agent.	除了大型语言模型, 现在处理整个代理。
189	We'll refer back to this slide later, but remember this.	我们稍后再参考这张幻灯片,但记住这一点。
190	The agent has to think, the agent has to think, and that's where the model comes in.	代理商必须思考,代理商必须思考,这就是模型出现的地方.
191	Perception, reasoning, planning, and using tools.	观念,推理,规划,以及使用工具.
192	All of that runs on the large language model, and the models are designed to do that.	所有这些都运行在大型语言模型上,模型都是为此设计的.
193	But the agent also has what is called a harness.	但特工也拥有所谓的绳索.
194	The harness is basically the exoskeleton of the digital agent.	牵引装置基本上是数码代理机的外骨架.
195	Just like we have the brain, the rest of our body represents our exoskeleton.	就像我们有大脑一样, 我们身体的其余部分代表我们的外骨骼。
196	The agent is the exoskeleton that orchestrates the work,	经纪人是负责组织工作的外骨骼
197	that allows the agent to actually engage memory, active memory, working memory,	允许代理人实际使用内存,活动内存,工作内存,
198	as well as long-term memory, and activating tools.	以及长期记忆和激活工具
199	Then the third part is a computer that processes knowledge,	第三部分是处理知识的计算机
200	trained for data processing.	接受数据处理培训。
201	And then lastly, a computer dedicated to CUDA-X libraries,	最后是一台CUDA-X图书馆的电脑
202	and its job is to run tools.	它的工作是运行工具。
203	And these computers now represent the entire agentic system.	这些计算机现在代表了整个代理系统。
204	So Vera Rubin is in full production, and it's nice to kind of see it.	所以薇拉·鲁宾已经完全投入生产了,也很高兴能看到它。
205	This is the first computer in the world that we've created.	这是我们创造的世界上第一台电脑。
206	No cables, no wires, completely 100% liquid-cooled, and it's just really delightful.	没有电缆,没有电线, 完全100%的液体冷却, 它只是真的很令人愉快。
207	Hang on.	坚持住。
208	Really delightful to be able to pull it out.	能把它拔出来真令人高兴
209	这是一台Vera Rubin电脑。
210	Okay, so this has two GPUs.	好吧,这有两个GPU。
211	Yep.	对
212	What used to take, in the case of Grace Blackwell, two hours to assemble one node.	在Grace Blackwell案中过去需要两个小时来组装一个节点
213	And when you see it, you'll know what I mean.	当你看到它,你会明白我的意思。
214	Cables and wires everywhere, 80% liquid-cooled.	到处都是电缆和电线,80%的液冷.
215	Now Vera Rubin is 100% liquid-cooled, and it takes less than five minutes to assemble it.	现在Vera Rubin是100%的液体冷却剂,组装它需要不到5分钟.
216	And so, incredible, incredible achievement.	所以,令人难以置信,令人难以置信的成就。
217	Of course, this translates to higher cycle times in manufacturing,	当然,这意味着更高的周期制造业,
218	much, much better reliability, and of course, a lot better performance.	可靠性更高,当然也更好
219	This is the NVLink switch tray, and what this does is it connects 72 of these GPUs.	这是NVLink的开关托盘,它能连接72个这些GPU.
220	There are four of them in here right now.	现在这里有四个人
221	It connects 72 GPUs into one gigantic GPU.	它将72GPU连接为一个巨大的GPU.
222	Okay, so each GPU is 20 petaflops, connecting 72 of them into one gigantic system.	好吧,所以每个GPU是20 petaflops, 连接其中72个到一个巨大的系统。
223	Okay.	摆
224	Thank you.	谢谢
225	And this is our new Vera computer.	这是我们的新Vera电脑。
226	It's okay.	没关系。
227	Go easy.	别紧张
228	Let her, give her a chance.	让她,给她一个机会。
229	Everybody has a chance to say something.	每个人都有机会说点什么
230	But today, let me, no, no, take, just, I'll talk to you later.	但今天,让我,不,不,拿,只是,我会跟你谈谈。
231	I'll talk to you guys later.	我待会再跟你们谈
232	Thank you.	谢谢
233	All important issues.	所有重要问题。
234	But let me get through my keynote.	但让我通过我的主旨发言。
235	We have, we have, so this, this auditorium is packed.	我们有,我们有,所以这个, 这个礼堂是收拾好的。
236	We have a hundred other watch parties around the world.	我们在世界各地还有一百个监视聚会。
237	A hundred watch parties.	一百个看台派对
238	15,000 people are watching us right now.	一万五千人在监视我们
239	And so, I'll get through the keynote.	因此,我会通过主旨。
240	There's some really, really fun things I want to show you in the end.	最后我想给你们看一些非常有趣的东西。
241	Okay, so you just have to, you have to, I have to torture you through some education first.	好吧,所以你只需要,你必须, 我必须折磨你通过一些教育先。
242	And then the entertainment is coming.	然后娱乐即将到来.
243	This, this our brand new, this is our brand new Vera computer.	这是全新的 Vera电脑
244	Vera CPU是一种非凡的CPU兽.
245	Eight instructions per clock.	每时八指示.
246	It has the world's first LPDDR memory system so that we can have high performance, very high bandwidth,	它拥有世界上第一个LPDDR内存系统,这样我们就可以有高性能,非常高的带宽,
247	as well as extremely excellent energy efficient, efficiency.	以及极其出色的节能和效率。
248	The on chip fabric is the fastest of any chip in the world.	芯片织物是世界上最快的芯片
249	As a result, data processing is off the charts.	因此,数据处理不在图表之列。
250	Agentic sessions starting and processing is off the charts.	启动和处理的代理会话不在图中。
251	And this is designed for doing exactly that.	这是完全为了做到这一点。
252	This is our Vera computer.	这是我们的Vera电脑。
253	And this computer here, this computer here, is the Spectrum X.	这台电脑,这台电脑,就是光谱X。
254	This is the world's first co-packaged optics.	这是世界上首个共同包装的光学仪器.
255	And you can see it here, I'll show you.	你可以看到它在这里,我给你看。
256	Some of this, some of this area falls into the ground.	其中有的,有的落在地上.
257	And some of this area you can walk on.	有些地方你可以继续走
258	I'm just not sure which.	我只是不确定是哪个
259	Alright, so, so if I accidentally disappear, you guys know, you guys know us.	好吧,所以,所以,如果我不小心消失, 你们知道,你们认识我们。
260	It's not a stunt.	这不是一个特技。
261	It's an accident.	这是意外
262	And so you can see here, these are the lasers.	你可以看到,这些是激光。
263	And these lasers come up through optics and it directly connects, directly connects to our chip.	这些激光通过光学技术直接连接到我们的芯片
264	And it connects using a process technology we invented with TSMC called COOP.	它使用我们发明的工艺技术和叫做COOP的TSMC连接起来。
265	And so silicon photonics are directly attached right onto our chip.	硅光子直接附着在我们的芯片上
266	Otherwise, what you would see is all of this would be PC board.	否则,你会看到这一切将是PC板。
267	And PC board's transmission capability is much worse than optical cables.	而PC板的传输能力比光学电缆要差得多.
268	And so we now have extremely high data rate, extremely high quality data that comes directly onto the chip.	因此,我们现在有非常高的数据率,非常高的质量数据直接出现在芯片上。
269	And this goes directly into, otherwise transceivers, without going through a PCB and a transceiver goes right into the optics.	这直接进入,否则就是收发机, 不经过PCB,收发机直接进入光学。
270	Okay, so this is the world's first CPO.	好吧,所以这是世界上第一个CPO。
271	We're in complete volume production.	我们正在生产完整的量。
272	And 512 port at 200 gigabits per second.	512个港口每秒200千兆比特
273	The world's most advanced Ethernet switch.	世界最先进的以太网开关.
274	And we're really proud of it.	我们对此非常自豪。
275	Thank you.	谢谢
276	And so, so what that, what that gives us, what that gives us is the ability to process the thoughts, the large language model.	所以,这给了我们什么, 给了我们什么, 给我们处理思想的能力, 大语言模型。
277	Process the agent harness.	处理代理控制器。
278	Run data processing for its knowledge.	运行数据处理以获取知识。
279	And then run accelerated computing for all of the tools.	然后对所有工具进行加速计算。
280	That represents essentially the agent workload.	这主要是代理工作量。
281	And I'll show you more in a second.	我一会儿再告诉你更多
282	While Vera Rubin is in full production and the performance is off the charts.	Vera Rubin正在全面制作中,而表演则不尽人意。
283	Every single generation, as I mentioned, through extreme co-design, we're able to achieve performances that are well beyond reason.	正如我所提到的,每一代人,通过极端的共同设计,我们都能实现远远超出理性的表演.
284	Well beyond what Moore's law would have dictated.	远远超出了摩尔定律的要求
285	And as you know, Moore's law has really largely ended.	如你所知,摩尔的法则已经基本结束了。
286	And so we need these extreme co-design ideas.	所以我们需要这些极端的共同设计的想法。
287	Working from chips to systems, systems to system software to algorithm to the entire AI factory.	从芯片到系统,系统到系统软件到算法到整个AI工厂的工作.
288	Well, Vera Rubin is another giant step on top of Blackwell.	Vera Rubin是布莱克威尔的又一个巨型台阶
289	Blackwell was about 30 times faster than Hopper in doing agentic, in doing large language model processing.	布莱克韦尔在做代理,做大型语言模型处理方面比霍伯快约30倍.
290	And now Vera Rubin is another 30 times greater than Blackwell in just two years time.	而现在薇拉·鲁宾在短短的两年时间里又比布莱克韦尔大30倍.
291	Let's take a look at how Vera Rubin turns into what is a DSX full stack AI factory.	让我们来看看Vera Rubin是如何变成 DSX全堆AI工厂的。
292	AI is creating the largest infrastructure build out in history.	大赦国际正在创建历史上规模最大的基础设施。
293	The factories are massive.	工厂是巨大的。
294	The investments are significant.	投资数额很大。
295	We have to build it right.	我们必须正确地建造它。
296	But great chips alone don't make a great AI factory.	但仅靠伟大的芯片并不能制造出一个伟大的AI工厂.
297	NVIDIA DSX is a full stack AI factory reference architecture, co-innovated with the world's leading infrastructure companies.	NVIDIA DSX是一个完整的堆栈AI工厂参考建筑,与世界领先的基础设施公司共同创新.
298	Compute, networking, storage, software, power and cooling, co-designed as one system.	计算,联网,存储,软件,动力和冷却,共同设计成一个系统.
299	Optimized from design to operation, DSX starts in a digital twin.	从设计到操作的优化,DSX从数字双胞胎开始.
300	Performance, power, water and scale validated virtually before construction begins.	性能、动力、水和规模几乎在施工开始前就得到验证。
301	PTC turns equipment designs into simulation-ready assets.	PTC将设备设计转化为模拟准备资产.
302	DASO Systems connects rack and facility design through model-based systems engineering.	DASO Systems通过基于模型的系统工程连接机架和设施设计.
303	Schneider Electric模拟电力分配和液体冷却.
304	Jacobs brings it all together in the omniverse digital twin.	雅各斯将它汇集在全数数码双胞胎中.
305	Siemens evaluates thermal, acoustic and grid impacts.	西门子对热、声学和电网影响进行评价。
306	And Procore supports construction phase planning.	而Procore则支持建设阶段规划.
307	Power limits how much an AI factory can produce.	动力限制AI工厂能生产多少.
308	DSX Max LPS is an end-to-end control loop that dynamically steers power to match workload demand.	DSX Max LPS是一个端对端的控制循环,动态引导电源以匹配工作量需求.
309	It uses 30% less power for the same output.	它对相同的输出使用30%的功率.
310	DSX rethinks cooling.	DSX重新思考冷却.
311	NVIDIA Vera Rubin systems are designed for closed-loop liquid cooling with coolant temperatures up to 45 degrees Celsius.	NVIDIA Vera Rubin系统是为闭路液冷却而设计的,冷却剂温度最高可达45摄氏度.
312	In suitable climates, outdoor dry coolers release heat into the air without evaporating water.	在合适的气候中,室外干燥的冷却器释放热量到空气中而不蒸发水.
313	At 2 gigawatts, DSX can save 54 million liters of water a day.	在2千兆瓦时,DSX每天可以节省5400万升的水.
314	That's the daily water use of 180,000 households.	这就是18万户人家的日常用水.
315	But an AI factory can take several months to build.	但AI工厂需要几个月的时间才能建成.
316	Until it is operational, it produces no tokens.	在投入使用之前,它不会产生任何符号.
317	Today, we are announcing NVIDIA DSX pod, bringing this full-stack platform into a factory-built, integrated system.	今天,我们宣布NVIDIA DSX舱, 将这个完整的平台带入一个工厂建造的集成系统。
318	它从Vera Rubin NVL72开始.
319	连接17公里Spectrum X电缆的7个液冷计算机架组成了一个Vera Rubin DSX舱.
320	Vera Rubin DSX pods are integrated, fully cabled and factory tested, then delivered in transportable containers.	Vera Rubin DSX舱室进行集成,全线缆和工厂测试,然后用可运输容器交付.
321	Dry coolers replace traditional evaporative cooling.	干冷却器取代传统的蒸发式冷却.
322	800 volts DC connects to the racks for efficient power delivery.	800伏特DC连接到机架以高效供电.
323	Once site utilities are ready, estimated on-site deployment goes from several months to days.	场地设施一旦准备就绪,现场部署估计从几个月到几天不等。
324	NVIDIA DSX AI Factory通过设计负责.
325	And now with DSX pod, AI factories can be built at the speed of light.	而现在有了DSX舱,AI工厂可以以光速建造.
326	More intelligence per watt.	每瓦增加情报
327	Less time to first token.	少一点时间到第一标语。
328	What do you think?	你觉得呢?
329	So, in 2016, we introduced the world's first AI supercomputer.	因此,在2016年,我们推出了世界上第一台AI超级计算机.
330	It was in one node, one GPU node.	它在一个节点, 一个GPU节点。
331	三年前,我们引进了Grace Blackwell NVLink72 架式计算机。
332	Both times, we had to reinvent not just the computer, the software, and how it works, but also the supply chain.	两次,我们不仅需要重塑电脑,软件, 以及它的运作方式, 而且还需要重塑供应链。
333	In the case of NVLink72, we took the entire rack, which would have otherwise been assembled on-site, and we assembled it inside the supply chain.	以NVLink72为例,我们拿走了整个架子,否则这些架子本会就地组装,然后在供应链内组装.
334	We essentially manufactured supercomputers at scale.	我们基本上制造的超级计算机规模。
335	It is the reason why we're able to move so fast and be able to provide for literally several hundred data center build-outs around the world at the same time.	这就是为什么我们能够快速移动并且能够同时提供几百个数据中心的建设。
336	Every single computer manufacturer in the world now manufactures NVLink72.	现在世界上每个电脑制造商都制造NVLink72.
337	The scale at which we are now producing computing is unheard of ever.	我们现在生产计算机的规模是前所未有的。
338	Unimaginable.	不可思议.
339	Well, even then, when you bring these systems onto your sites, there are still months and months of assembly.	即便如此,当你把这些系统带到你的网站上, 仍然有几个月和几个月的组装。
340	And you just heard 17,000 kilometers of wiring to connect just spectrum X for the scale-out computing.	你刚刚听到17,000公里的线条用来连接X频谱用于规模的计算
341	Seventeen kilometers.	17公里长.
342	Seventeen thousand kilometers.	17000公里长.
343	It's a crazy number.	这是一个疯狂的数字。
344	And so, 17 kilometers.	因此,17公里。
345	Incredible numbers.	不可思议的数字。
346	Now, suppose we were to do that all inside the supply chain.	现在,假设我们这样做全部在供应链。
347	Run everything up, build everything up, make sure that it runs completely, get it qualified, burnt in, and by the time we ship it, it is fully functional.	运行所有东西,建立所有东西,确保它完全运行,获得合格,燃烧,到我们发货时,它已经完全运转.
348	The amount of cabling and on-site assembly is going to be greatly reduced.	电缆和现场装配的数量将大大减少。
349	There's something much, much more.	还有很多事情
350	In fact, what's going to happen here is that by moving this into the supply chain, the amount of idle computing, the amount of idle computing that's going to be sitting inside at your data centers will reduce dramatically.	事实上,这里会发生的是, 通过将这个移动到供应链中, 闲置的计算数量, 闲置的计算数量, 将会在你的数据中心里大量减少。
351	Remember this.	记住这一点。
352	Each gigawatt, each gigawatt data center is $50 billion.	每千兆瓦,每千兆瓦数据中心为500亿美元.
353	The revenue generation you expect is tens of billion dollars per year.	你所期望的年收入是数百亿美元。
354	And so, we just said it's $40 billion per year of revenue generation that comes out of that infrastructure.	因此,我们刚刚说,每年从基础设施中产生400亿美元的收入。
355	$40 billion, if delayed by just three months, that's $5 billion.	400亿美元,如果推迟了三个月,那就是50亿美元。
356	Excuse me, what is that?	对不起,那是什么?
357	$10 billion worth of revenues.	收入100亿元.
358	$10 billion worth of revenues will make anybody anxious.	价值100亿美元的收入将使任何人感到焦虑。
359	And so, by moving this into the supply chain, the speed by which we can turn on, go from installation, arrival of computing, to production of tokens will go from many months, we think something along the lines of four to five months down to about a month.	因此,通过将它转移到供应链中, 我们可以打开它的速度, 从安装,计算机的到来, 到制作信物, 将会从许多月, 我们想到一些东西, 从4到5个月到大约1个月。
360	Okay, so this is a complete transformation of the world supply chain, and this will change the way that people install and build data centers.	好了,这就是世界供应链的彻底转变, 这将改变人们安装和建设数据中心的方式。
361	And hopefully, as a result, velocity goes up, the wasted idle compute, the wasted idle CAPEX will reduce, and everybody becomes more profitable.	希望,因此,速度会上升, 浪费的闲置计算, 浪费的闲置CAPEX会减少, 每个人都会更赚钱。
362	And so, this is MVLink 72 now into a large cabinet system.	所以,这是MVLink 72 现在进入一个大型的内阁系统。
363	We call it DSX pod.	我们称之为DSX舱。
364	Okay, so this is the next click of NVIDIA computing.	好了,这是NVIDIA的下一击。
365	Now, what you make and how you make it makes you special.	现在,你做的是什么,你是如何让你特别的。
366	Just as here in Europe, you, the work you know makes you special.	就像在欧洲,你,你知道的工作让你特别。
367	You design machines, you operate factories, you discover medicine, you have extraordinary craft.	你设计机器,经营工厂,发现医学,拥有非凡的工艺.
368	That craft, that specialty makes you special.	那个手艺,那个专业让你特别
369	Well, what makes NVIDIA special are really two things.	那么,是什么使得NVIDIA特别是真正的两件事。
370	The first thing is that our architecture is the only accelerated computing architecture that is general purpose.	首先,我们的建筑是唯一一个通用的加速计算建筑。
371	We don't specialize on just one algorithm, for example, transformers.	我们不专门研究一个算法,例如变压器。
372	We specialize across anything that could be paralyzed.	我们专门研究任何可能瘫痪的东西。
373	Could be fluids and particles, it could be quantum, it could be quantum chemistry, could be image processing, could be data processing, structured vector data processing.	可以是流体和粒子,可以是量子,可以是量子化学,可以是图像处理,可以是数据处理,是结构化的矢量数据处理.
374	Processing of all kinds.	各种处理.
375	And because we've been working at it for so long, and because NVIDIA's architecture has remained consistent for literally two and a half decades.	因为我们工作了这么长时间, 并且因为NVIDIA的建筑已经维持了25年。
376	The number of algorithms, the number of domains, the number of industries we serve, we accelerate, has grown tremendously.	算法的数量,域的数量, 我们服务的产业的数量,我们加速, 已经取得了巨大的发展。
377	Our install base, our general purpose architecture, our dedication to this architecture, and our collaboration with all of the world's industries	我们的安装基地,我们的通用建筑,我们对这一建筑的奉献,以及我们与世界所有工业的合作
378	has made NVIDIA's architecture literally everywhere.	已经使NVIDIA的建筑几乎无处不在。
379	The second thing is how we do it.	第二件事是我们怎么做到的。
380	Of course, we have the architecture that we developed, but we also build it now in a full stack, complete AI factory.	当然,我们有我们开发的建筑, 但我们现在也把它建在一个完整的堆栈,完整的AI工厂。
381	There are many companies in the world who want to design their own chips.	世界上有很多公司想自己设计芯片.
382	But most of the companies in the world just want to deploy the AI infrastructure so that they can run their business on top of it.	但是,世界上大多数公司只是想部署AI基础设施,以便他们能够经营自己的业务。
383	The ability to design a chip and build a supercomputer, write all the software, is really completely out of mind for most of the companies around the world.	设计芯片和建造超级计算机的能力,写出所有软件的能力,对于世界上大多数公司来说,其实是完全疯了.
384	And so we now have the combination of these two ideas, CUDA, full stack AI factories, and it exposes us to markets that otherwise isn't possible.	现在我们有了这两个想法的结合, CUDA, 完整的堆积AI工厂, 它让我们暴露在市场中, 否则是不可能的。
385	We're able to partner with so many companies here for this very reason.	为了这个原因,我们能和这么多公司合作
386	If I were to go around this, starting with CSPs, this is where most people think the vast majority of AI is today.	如果我绕过这个,从CSP开始,这是大多数人认为今天绝大多数AI的地方.
387	But as you can see in the beginning of our keynote, I mentioned a whole bunch of companies who are not those.	但正如你们在主旨发言开头所看到的, 我提到了一大堆公司,它们不是那些公司。
388	So CSPs remain very large businesses for the world's computing.	因此CSP仍然是世界计算学的非常大的企业.
389	But there are AI natives, companies that are starting all over the world, there are so many that are here,	但有AI本地人, 公司开始在世界各地, 有这么多在这里,
390	Neoclouds, because every region wants to participate in the AI economy.	Neoclouds,因为每个地区都希望参与AI经济.
391	Every region wants to ensure that the valuable resources that they have, the energy that they have, is dedicated to companies in their region.	每一个区域都希望确保它们所拥有的宝贵资源,以及它们所拥有的精力,都献给本地区的公司.
392	For example, Mistral here in Europe, Mistral in France, Nebius here in Europe, so many companies around Europe that are building up Neoclouds.	例如,欧洲的米斯特拉尔,法国的米斯特拉尔,欧洲的尼比乌斯, 如此多的欧洲公司正在建立新星。
393	OEMs and ODMs who are building computers for the world's enterprise.	为世界企业建造计算机的OEMs和ODMs.
394	AI labs, of course we know about last year, all we spoke about was OpenAI.	AI实验室,我们当然知道去年,我们谈论的只是OpenAI。
395	今天我们谈论OpenAI,Anthropic, 但我们也谈论不可思议的Metasmuse, Grokbot, 当然还有双子座。
396	So many different AI labs are now participating at the frontier.	现在有很多不同的AI实验室在边境参与.
397	And then very importantly, and this is ultimately the most important,	这才是最重要的
398	for AI to succeed, for Europe to benefit from AI and to have really created an AI ecosystem and AI economy,	让AI成功,让欧洲从AI中获益,真正创造了AI生态系统和AI经济,
399	you need all of the end markets, the end industries to adopt AI and to succeed with AI.	您需要所有终端市场,终端行业采用AI,并成功AI。
400	Whether it's in the automotive industry, because we're building self-driving cars, or the way you manufacture with robotics,	无论是在汽车工业, 因为我们正在建造自驾汽车, 还是你制造机器人的方式,
401	the way you discover new drugs in healthcare, your financial services industry,	你发现新药物的方式在医疗保健,你的金融服务行业,
402	all of these different industries have to become AI companies, every single one of them.	所有这些不同的产业都必须成为AI公司,每一个都成为AI公司.
403	Well, we have a unique ability through what I just mentioned about CUDA and the way that we build computers.	通过我刚才提到的CUDA和我们制造计算机的方式,我们拥有独特的能力。
404	We have a unique ability to serve literally every single segment of AI.	我们拥有独特的能力,从字面上为AI的每一个部分服务.
405	And so one architecture, many markets, and because these infrastructures are so significant and so expensive,	因此,一个建筑,许多市场, 由于这些基础设施是如此重要如此昂贵,
406	we need to make sure that they're as productive as possible, as soon as possible,	我们需要确保它们尽可能有生产力尽快
407	that they're fungible, that you can use it for every workload, from the beginning of your AI journey,	他们可以互换, 你可以使用它每一个工作量, 从开始你的AI旅程,
408	data processing to training, to post-training, inference and agentic processing, every single step of the AI journey,	数据处理到培训后培训推论和代理处理
409	to every type of AI, chemicals and physics, to languages, to every single industry and the algorithms that they use.	各种类型的AI,化学和物理,语言, 每一个行业和它们使用的算法。
410	And so NVIDIA's architecture is fungible, and as a result of that, continuous innovation, continuous updating of software,	因此NVIDIA的架构是可互换的,因此,不断创新,不断更新软件,
411	because the install base is so great, we have such an incredibly rich ecosystem of software,	因为安装基础如此之大我们拥有如此丰富的软件生态系统
412	then these architectures are also durable, fungible, durable, productive.	那么这些建筑也是耐久的,可替换的,耐久的,有生产力的.
413	Well, that lays the foundation.	这奠定了基础。
414	That is the infrastructure.	这就是基础设施。
415	Now what do we do with it?	现在怎么办?
416	This last year, we saw a transition in AI.	去年,我们看到了AI的转变.
417	AI is now productive.	AI现在很有成果。
418	And one of the most important times was in the beginning of the year when we spoke about OpenClaw,	其中最重要的一次是在年初我们谈到OpenClaw时
419	which introduced the concept of an open source agent, and it really started what we all see across the world today.	它引入了开源代理的概念, 它真正地开始了我们今天所看到的。
420	Agents think, reason, iterate, use tools, and as a result, consume an enormous amount of tokens.	特工们思考,理性,仪表,使用工具,结果消耗了大量的代币.
421	But as a result, they become useful.	但因此,它们变得有用.
422	So more agents are being developed, they're more useful,	所以,更多的代理正在开发, 他们更有用,
423	and therefore they're used by more people, and because they have to use a lot of tokens to think,	因此他们被更多的人使用因为他们必须用很多符号来思考
424	the amount of token consumption has gone through the roof.	象征性消费的数量已经从屋顶上走过。
425	GPU demand has really gone through the roof.	GPU的需求真的突破了屋顶.
426	Now what's processing it is this.	现在处理的是这个。
427	This is an agent.	这是特工
428	These are the five elements of an agent.	这些都是五要素一个代理。
429	The first part are the language models.	第一部分是语言模型。
430	These are the tokens that are running on top of the GPUs,	这些是运行在GPU上面的标志,
431	and they're generating tokens that are consumed by the agent, what I call the exoskeleton.	他们正在生成代币被毒剂消耗, 我称之为外骨骼。
432	That OpenClaw harness sits inside a sandbox, and that sandbox isolates it from the rest of the system.	OpenClaw的绳子坐落在沙盒里沙盒将它与系统的其他部分隔离开来
433	This is a very important concept that we're going to come back to.	这是一个非常重要的概念,我们将回到。
434	There are five things.	有五义.
435	There's the model, there's the harness, the sandbox that the harness sits in,	有模型,有绳子, 沙盒,绳子坐在那里,
436	and two other things, knowledge and tools.	还有另外两件事知识和工具
437	Models, sandbox, harness, knowledge and tools.	模型、沙盒、绳索、知识和工具。
438	These are the five elements of an agent.	这些都是五要素一个代理。
439	Now before I talk about how to make the agents useful,	在我讲到如何让特工有用之前
440	the most important thing is we have to make sure that the agent is secure.	最重要的是我们必须确保特工安全
441	The agent is autonomous in this autonomous system that has access to files, has access to resources, networks,	代理在这种自主系统中是自主的,可以访问文件,可以访问资源,网络,
442	access to tools, and we need to make sure that the privileges that we provide this autonomous system is minimal	获取工具,我们需要确保我们提供的这种自主系统的特权是最低限度的
443	and that we constantly monitor it.	并且我们一直监视着它
444	We introduced a fantastic new product, a new platform, and it's called OpenShell.	我们推出了一个奇妙的新产品,一个新的平台,它叫做OpenShell.
445	Openshell来得正好
446	This is a secure runtime that provides for the things that I just mentioned	这是一个安全的运行时间提供的东西我刚才提到
447	and has an additional hardware-based monitor.	并拥有额外的基于硬件的显示器.
448	Before I explain it to you, let's watch a video.	在我给你解释之前,让我们看看视频。
449	Prizes are preparing to deploy billions of AI agents.	奖项准备部署数十亿AI特工.
450	To be useful, they'll reason, plan, use tools, access data, and work together.	为了有用,它们会理性,计划,使用工具,访问数据,并合作.
451	Today, the only thing stopping an agent from misusing all that is asking it nicely.	今天,唯一能阻止特工滥用一切的就是善意的要求
452	Please don't delete anything.	请勿删除任何东西。
453	Please don't push to main.	请不要推到主。
454	Please don't touch my SSH keys.	请不要碰我的SSH钥匙。
455	That's the please don't approach.	请不要靠近
456	But hope isn't security.	但希望不是安全的。
457	Safety isn't an aspiration.	安全不是一个愿望。
458	It's technology and engineering.	科技与工程.
459	NVIDIA Open Shell和哨兵将这一哲学付诸实践。
460	You write a policy for what each agent can touch, and OpenShell enforces it at the granularity of system calls, file access, and network connections.	您为每个代理能够触摸的东西写一个策略, OpenShell 在系统调用、文件访问和网络连接的颗粒性上执行它。
461	Formal verification checks every policy change against your permission ceiling, so agents never gain more access than intended.	正式核查根据你的许可上限检查所有政策变动所以特工们从来就得不到比预期更多的访问机会
462	It's like the early internet.	这就像早期的互联网。
463	Viruses were everywhere until browsers enforced strict limits on what a web page could do.	病毒无处不在,直到浏览器对网页所能做的事情实施严格的限制.
464	OpenShell对特工也是这样
465	Enforcement happens outside the agent, not just in LLM guardrails, bringing agent-native security into your infrastructure.	执法发生在特工之外不只是在LLM守卫设施里给您的基础设施带来代理安全
466	Every agent runs inside its own isolated sandbox, with no direct route to the network.	每个特工都运行在自己孤立的沙盒内,没有直接通往网络的路线.
467	Every outbound network request goes through the supervisor, outside the sandbox.	每一个外出网络请求都通过主管,在沙盒之外.
468	It checks the action against your policy.	它检查行动与你的政策。
469	If it's not permitted, it's denied.	不许者,即绝也.
470	Keys stay outside, with the supervisor, and are added only on the way to an approved endpoint.	钥匙留在外面,与主管一起,只在经过批准的终点处的路上才增加。
471	A compromised agent can't leak a key it never had.	失密的特工不能泄露它从来没有过的钥匙
472	The gateway is the control plane.	网关是控制飞机.
473	The supervisor makes the decisions.	由主管决定。
474	The sandbox holds the agent, and the policy prover checks proposed policies against your permission ceilings.	沙盒持有代理, 政策证明人检查拟议的政策与您的许可上限。
475	OpenShell combines Linux kernel isolation with runtime policy enforcement.	OpenShell将Linux内核隔离与运行时政策执行相结合.
476	NVIDIA Sentry adds an optional layer of protection outside the host itself, away from the agent.	纳米比亚哨兵在宿主本身之外,远离代理人,增加了一个可选的保护层.
477	Running on NVIDIA BlueField 4, Sentry monitors agent activity and enforces security policies in a hardware-isolated trust domain.	在NVIDIA Blue Field 4上运行,哨兵监视代理活动,并在硬件同化的信托域执行安全政策.
478	Because agent reasoning traffic passes through Sentry, if the agent is planning to do something unintended, it will update policies immediately and cut the connection to the model, quarantining it in milliseconds.	因为代理推理流量通过哨兵,如果代理正在计划做一些意料之外的事情,它会立即更新政策,并切断与模型的连接,以毫秒的速度断绝.
479	And it keeps tamper-evident records that operators can verify for detection, investigation, and response.	它不断篡改不言自明的记录操作人员可以核实用于侦测、调查和反应
480	OpenShell is open source under Apache 2.0, runs everywhere your agents do, edge to cloud, and takes two commands to adopt.	OpenShell在Apache 2.0下是开放源代码,运行到您的代理处,边缘到云,并需要两个命令才能通过.
481	Install the CLI, create a sandbox, and run your favorite agent.	安装 CLI, 创建一个沙盒, 运行您最喜欢的代理。
482	Agents are only as useful as the access you give them.	特工们和你给他们的准入一样有用
483	OpenShell and Sentry put that access under your control.	OpenShell和哨兵把进入权交给你
484	That's how we advance AI capabilities and safety together.	这就是我们一起推进AI能力和安全的方法.
485	Agents are autonomous systems.	特工是自动系统
486	We have to create extremely secure technology to make sure that we could develop these agents and deploy these agents safely.	我们必须建立非常安全的技术,以确保我们能够发展这些代理人,安全部署这些代理人。
487	Now, it's incredible.	现在,这是不可思议的。
488	After announcing OpenShell, just about every major company in the world, from the Frontier Labs to the operating system companies, cybersecurity companies, platform companies, have come out to adopt OpenShell.	在宣布OpenShell之后,仅仅世界上每个大公司,从边境实验室到操作系统公司,网络安全公司,平台公司,都出来采用OpenShell.
489	It's really fantastic to see.	这真是太棒了。
490	And the reason for that is because this technology has arrived at exactly the right time.	之所以如此,是因为这个技术已经到达了正确的时机。
491	At a time when we want to deploy agents all over the world.	此时我们想在世界各地部署特工。
492	Now remember, this isn't just about the services using OpenShell as they deploy agents.	现在记住,这不仅仅是服务使用OpenShell 他们部署代理。
493	This is really about all of the world's companies also adopting OpenShell so that you could receive these agents and have them run autonomously inside your company.	这是关于世界上所有的公司也采用OpenShell 这样你就可以接收这些代理让他们在你的公司里自主经营。
494	Every single level of capability, every single task that you give the agent for every person that uses the agent, every one of those combinations requires a different level of privilege that you would provide to the agent.	每一级的能力,你给代理人的每一项任务每一个使用代理人的人,每一个这些组合都需要你给代理人提供不同的特权
495	And every single time an agent is launched, the privileges are different.	而每次推出代理时,特权都不同.
496	And so privileges, the policy enforcement, the monitoring, all has to be done dynamically.	因此,特权,政策执行,监控, 都必须动态地进行。
497	所以OpenShell是真正的革命。
498	I'm delighted to say that Anthropic and NVIDIA are partnering to create cloud OpenShell agents.	我很高兴地说,Anthropic和NVIDIA正在合作创建云开壳代理.
499	And so you can go to the website.	然后,你可以去网站。
500	There's a really great write-up about that.	写得很好
501	所以,第一部分,第一部分是如何使用这些带子, OpenClaw, Hermes, OpenHand, 你可以从Langchain, DeepAgents, 或者用Python来写。
502	There's a whole bunch of different ways you could create agent harnesses that sits inside now an OpenShell secure harness.	有一大堆不同的方式你可以创造特工的带子现在坐在里面一个OpenShell安全带。
503	Okay, and that's the sandbox that it sits in, the secure sandbox.	好吧,这就是它坐在那里的沙盒,安全沙盒。
504	And then next, let's talk about models.	然后,我们来谈谈模型。
505	You should use models of all kinds.	你应该用各种模型
506	In the future, it won't just be one model that we run everything on.	在未来,它不会只是一个模型我们运行的一切。
507	We're going to run, we're going to use the best models whether they're closed, open, or custom.	我们要运行,使用最好的模型无论它们关闭,开放,还是习惯。
508	We're going to use all the models to do your work.	我们要用所有的模型来完成你的工作。
509	Sometimes the work that you do is fairly generic.	有时候你的工作很普通
510	Artificial general intelligence is going to do an incredibly good job.	人工一般情报工作会做得非常好
511	However, sometimes the work that you want to do captures too much of your proprietary knowledge, too much of your specialty, and you want to continuously enhance that intelligence.	然而,有时你想要完成的工作会捕捉到太多的专有知识,太多的专长,你想要不断增强这种智力.
512	Every company in the future will use both closed models as well as open models that you customize.	未来每个公司都会同时使用封闭模型以及您定制的开放模型.
513	By the end of the session, you will see that, in fact, building your own AIs is not that hard.	到会议结束时,你会看到,事实上,建立自己的AI并不是那么难.
514	It starts with, of course, the model.	当然是从模型开始的
515	We've been dedicated to advancing open models to open up new opportunities and enable people to build their own custom AIs.	我们致力于推进开放模式,开辟新机遇,使人们能够建立自己的定制AI.
516	Nemotron is for languages.	尼莫特龙是语言
517	Cosmos is our world foundation model.	宇宙是我们的世界基础模型。
518	We are the frontier.	我们是边界。
519	Groot for human robotics.	人类机器人的格鲁特
520	We are the frontier.	我们是边界。
521	Ising for autonomous navigation.	进行自主导航。
522	We are the frontier.	我们是边界。
523	Ising for quantum classical computing.	Ising用于量子古典计算.
524	We are the frontier.	我们是边界。
525	And, of course, Earth-2, physics, AI.	当然还有地球二号物理 AI
526	We are the frontier.	我们是边界。
527	So, in all of these different areas, we offer you all of the models, and we're continuously working on it.	因此,在所有这些不同的领域, 我们向你们提供所有的模型, 我们正在不断研究。
528	We've been working on it for a decade, and we'll continue to work on this forever.	我们为此努力了十年之久,我们将永远继续为此而努力.
529	We offer you the model completely open, the data that goes along with it so that you know the providence of the model that we created,	我们为你们提供了完全开放的模型, 随附的数据,让你知道我们创造的模型的本能,
530	as well as the scripts necessary for you to reproduce it.	还有你复制它所必需的剧本
531	So, not only do you have the open weights, you have from us the entire recipe and the data to recreate the open models	所以,你不仅有开放的权重, 你从我们那里得到了整个食谱和数据来重现开放的模型
532	so that you could know where it came from and have complete trust into it.	以便你们知道从哪里来的,并完全信任它。
533	Well, these models are not necessarily the frontier of frontiers, especially for the language model.	这些模型不一定是前沿的特别是语言模型
534	However, with customization and with partners that the ecosystem that we work with, you could turn these models into frontier.	然而,随着我们的生态系统的定制化和合作伙伴,你可以把这些模型变成前沿。
535	And, as a result, turn your proprietary data into sovereign intelligence.	结果,把你的专有数据变成主权情报
536	Here's an example of A21 helping Alice turn Nemotron into a model that exceeds frontier capabilities.	以下是A21帮助爱丽丝将尼莫特龙变成超越前沿能力的模型的例子.
537	Here's Base 10 helping CodeRabbit turning a Nemotron model, after fine-tuning on your data, turned into frontier capability, Base 10 and CodeRabbit.	这是10号基地帮助CodeRabbit转换一个Nemotron模型在微调你的数据后变成了前沿能力 10号和CodeRabbit
538	Here's Trajectory and Harvey, again using Nemotron, post-trained it and turned and created their own frontier-level capability.	这是Trajectory和Harvey,再次使用Nemotron,后训练它,并转身并创建了自己的前沿能力.
539	这是Skydance的Scale公司再次做同样的事情
540	It goes on and on and on.	随顺不断.
541	The point being that you could start with an open model, something that's incredibly good, like Nemotron,	重点是你可以开始一个开放的模型, 一些令人难以置信的好, 像尼莫特隆,
542	and with your own data, post-train it for your use case, and achieve, of course, super intelligence.	并且用你自己的数据, 后训练它为您的使用案例, 当然,实现超级智能。
543	That is a very sensible thing to do, and all the technology exists today.	这是一件非常明智的事情,所有的技术今天都存在。
544	Here's an example of Palantir with Tampa General Hospital.	以下是帕兰蒂尔与坦帕总医院的一个例子.
545	The list goes on and on.	名单不断.
546	We have a rich ecosystem of partners that can help you, and they're incredible companies, and they could work with you from around the world.	我们有一个丰富的生态系统合作伙伴可以帮助你, 他们是不可思议的公司, 他们可以与你合作从世界各地。
547	I'll mention one and I'll offend everybody else.	我会提到一个,我会冒犯其他人。
548	I'll have to sit here and mention all of them.	我不得不坐在这里提到他们的全部。
549	I love them all.	我爱他们
550	First, agent harness running in a secure sandbox open shell.	首先,特工的吊带运行在一个安全的沙盒打开壳。
551	Second, Nemotron large language models, the first two components of what you need.	第二,尼莫特隆大语言模型,你需要的前两个组件.
552	But ultimately, what turns these agents into super agents is that you give them access to your company's knowledge.	但归根结底,这些特工变成了超级特工就是你让他们了解你的公司
553	And this is really special.	这真的很特别。
554	This is called the Agent Knowledge Foundation that we put together.	这叫"特工知识基金会" 我们一起建立的
555	Libraries and models that are completely world-class.	图书馆和模型都是世界级的
556	It allows you to turn your company's data, the data could be in, the assets could be in unstructured format in documents,	它允许你调用你公司的数据, 数据可能是在, 资产可以是非结构格式的文件,
557	so that you need an AI to understand how to read these PDFs and read all these unstructured documents, or it could be in structured data inside databases.	因此,你需要一个AI来理解如何读取这些PDF并读取所有这些无结构的文档,或者可以在数据库内部的结构化数据中读取.
558	We have the libraries and the tools necessary to help you extract meaning and understanding from your company's data.	我们拥有必要的图书馆和工具来帮助你从公司的数据中获取意义和理解.
559	One at a time, Nemotron parse, completely world-class top of leaderboard.	一次一个,尼莫特龙完全世界一流的领导板。
560	Nemotron embed, taking a query and associating with the knowledge that's inside the vector database.	Nemotron嵌入, 接受一个查询并与在矢量数据库中的知识联系起来。
561	Top of leaderboard.	上头板.
562	Nemotron re-rank, it might find many different sources where the query is associated with different pieces of data.	Nemotron重新排序,它可能会发现许多不同的来源,其中查询与不同的数据块相关.
563	It brings it together, ranks them based on which one's the most relevant to you.	它将它聚集在一起, 排序他们的基础一个最适合你。
564	Kumo, to be able to predict from your structured database what is likely to happen.	库莫,从你结构化的数据库里能够预测会发生什么
565	And so this is, for example, if you were to receive a notice that one of your suppliers will be short on supply or delay on supply,	因此,举例来说,如果你收到一个通知, 你的供应商将短缺供应或延迟供应,
566	that piece of information is likely unstructured data as a document.	作为文档,该信息可能是非结构数据。
567	The information about the orders in your company are likely inside a structured database, your SQL database, for example.	有关您公司订单的信息可能存在于一个结构化的数据库,例如您的SQL数据库.
568	And in order to predict what's going to happen to all of these orders who are going to be short, the amount of delay, you use Kumo to make that prediction.	为了预测这些订单会怎样, 这些订单将会是短暂的, 延迟的数量, 你用Kumo来预测。
569	And then in order to figure out what to replace it with, you might use vector database search and go find an engineering document of a replacement component.	然后,为了找出用什么来替换它, 你可以使用矢量数据库搜索, 并去找一个工程文件一个替换组件。
570	Okay, so all of those different libraries and data processing capabilities now makes for you your knowledge foundation.	好了,所以所有这些不同的库和数据处理能力现在让你的知识基础。
571	You've now armed your AI with proprietary data, proprietary knowledge.	你现在已经武装你的AI 拥有专有数据,专有知识。
572	Now, the last part they'll talk about is ultimately tools.	现在,他们要谈的最后一部分是工具。
573	Tools of the trade.	交易工具。
574	NVIDIA has been working on CUDA-X, and for most of you, before AI, all of our conversations have been around how do we accelerate simulators and solvers,	NVIDIA一直在研究CUDA-X,对你们大多数人来说,在AI之前,我们所有的对话都围绕着我们如何加速模拟器和解析器,
575	data processing engines, domain-specific libraries of all kinds.	数据处理引擎,各类域特定库.
576	They're used in scientific discovery in all the different fields of science.	它们被用于所有不同科学领域的科学发现.
577	They're used in engineering.	它们被用于工程.
578	They're used in research.	它们被用于研究.
579	They're used in designing, building creative arts.	它们被用于设计,建设创作艺术.
580	At this core are CUDA-X libraries, and these CUDA-X libraries are algorithms.	在这个核心是CUDA-X库,这些CUDA-X库是算法.
581	And algorithms, as you know, are beautiful.	而算法,你知道,是美丽的。
582	Let's take a look at them.	让我们看看他们。
583	Math is beautiful.	数学很美丽
584	CUDA-X libraries, these are the crown jewels of our company, and the reason for that is this.	CUDA -X图书馆这是我们公司的皇冠珠宝原因就是这个
585	When you're building an AI assistant, it could be a lab assistant, it could be a research assistant, it could be a co-designer,	当你建造一个AI助理, 它可以是实验室助理, 它可以是研究助理, 它可以是联合设计师,
586	as somebody who's co-creator with you, building something with you.	作为你共同创作的人, 与你一起建造一些东西。
587	Ultimately, what makes those agents useful, what makes them able to do real work, are the tools.	归根结底,是什么使这些代理人有用,是什么使他们能够做真正的工作,就是这些工具。
588	And so we now have models that allow these agents to reason, understand information, reason, to be able to plan and iteratively act and use tools until it could get the job done.	因此,我们现在有了模型, 允许这些代理商理性,理解信息,理性, 能够规划和反复行动, 使用工具,直到它能够完成工作。
589	But it still comes down to the tools, and these tools we've been working on for so many years.	但是,它仍然下到工具, 而这些工具我们已经工作了这么多年。
590	And there was one particular computer I showed you on the right-hand side that runs every single one of these tools, and it could be fluid dynamics and particle physics.	我给你们看的右手边有一台电脑运行这些工具中的每一个, 它可以是流体动力学和粒子物理学。
591	It could be quantum chemistry.	可能是量子化学
592	It could be data processing.	这可能是数据处理。
593	It could be ray tracing.	可能是射线追踪
594	All of these tools run on that one computer, and that one computer speeds up the processing of it.	所有这些工具都运行在一台计算机上, 而一台计算机可以加速它的处理。
595	compared to just running it natively on a CPU,	而不是在CPU上用本地语言运行
596	由10,50,100x。
597	And CUDAX tools is what give the agents the tools of the trade.	而CUDAX工具是给予代理商交易工具的.
598	And it's so cool to see how other people are using it.	看别人怎么用它,太酷了
599	Let's take a look.	让我们来看看。
600	So this is the magic, magic behind the agents.	这就是特工背后的魔法
601	Ultimately, the agents have to use tools.	归根结底,特工必须使用工具。
602	Now, of course, in one of the first fields of agentic use	现在,当然,在最早的药剂使用领域之一
603	is software programming.	是软件编程。
604	And therefore, a compiler or a Python is all really it needs.	因此,编译器或Python是它真正需要的。
605	But in the future, as we have these tools do more and more	但在未来,我们有这些工具做越来越多的
606	and more and more useful work, we're	和越来越多的有用的工作,我们
607	going to need to give them more tools.	需要给他们更多的工具。
608	And here's some examples.	这里有一些例子。
609	So here's Lyta Sciences.	这就是莱塔科学。
610	They're creating a lab assistant.	他们正在建立一个实验室助理。
611	他们使用NVIDIA的BioNemo模型
612	that are designed for understanding proteins	用来理解蛋白质
613	and understanding cells and generating proteins	理解细胞和生成蛋白质
614	and chemicals in drug discovery.	和药物发现中的化学品。
615	他们使用BioNemo和CUDAX工具.
616	This is Siemens Factory, building a factory agent	这是西门子工厂,建造一个工厂代理
617	that uses NVIDIA Cosmos, our world foundation	我们的世界基金会
618	model that understands the physical world and CUDAX tools.	能够理解物理世界和CUDAX工具的模型。
619	And here's an example of Dassault	这是达索的一个例子
620	creating a virtual companion that	创建虚拟伴侣
621	使用NVIDIA的Nemotron, 我们的语言模型,
622	还有CUDAX工具
623	Let's see them in use.	让我们看看他们正在使用。
624	The most intelligent scientist still needs a laboratory.	最聪明的科学家仍然需要一个实验室.
625	The best engineer still needs simulators and solvers.	最好的工程师仍然需要模拟器和解析器.
626	An AI agent is no different.	人工智能特工也一样
627	An agent brings together five building blocks.	一个特工集中了五个构件
628	The model perceives reasons and plans.	模型能感知原因和计划.
629	The harness coordinates the work.	绳子协调工作
630	Knowledge foundation gives the agent context	知识基础提供了代理商的背景
631	from your company's structured and unstructured data.	从你公司的结构化和无结构化的数据。
632	Tools let the agent simulate, experiment, and act.	工具让代理模拟,实验,并采取行动.
633	NVIDIA Open Shell控制着哪些代理可以访问和执行.
634	Start with a frontier closed model	从前沿封闭模型开始
635	or customize an open model.	或自定义打开的模型。
636	With NVIDIA's agent toolkit, any company	NVIDIA的代理工具箱任何公司
637	can turn its knowledge and expertise	能够改变其知识和专长
638	into a customized super agent.	成为定制的超级特工
639	Let's look at three examples.	略举三例.
640	对Lyla Sciences来说目标是设计一种疗法
641	for B-cell lymphoma.	B细胞淋巴瘤
642	Lyla Iris的代理商在NVIDIA尼莫特龙之上
643	turns that goal into an experimental plan.	把目标变成实验计划
644	In the computational dry lab, Lyla agents	在计算干燥实验室,Lyla代理
645	use NVIDIA BioNemo Agent Toolkit	使用 NVIDIA 生物内莫剂工具包
646	to call models for protein design and RNA optimization,	可以调用蛋白质设计和RNA优化模型,
647	running millions of virtual experiments.	运行着数百万个虚拟实验
648	The most promising candidates move	最有前途的候选人流动
649	into Lyla's AI Science Factory,	进入莱拉的AI科学工厂,
650	where autonomous robots and instruments,	其中自主机器人和仪器,
651	由NVIDIA Omniversis和宇宙公司开发,
652	make and test them.	做测试
653	The results return to Iris, shaping the next designs	结果是返回到Iris,塑造下一个设计
654	and experiments.	和实验。
655	The loop continues.	循环继续。
656	At Siemens Electronic Factory Erlangen,	在西门子电子工厂埃兰根
657	the goal is to bring new products into production.	目标是将新产品投入生产。
658	Process preparation ex-agents use NVIDIA Cosmos	流程准备前代理人使用 NVIDIA 宇宙
659	to generate virtual prototypes and synthetic data,	生成虚拟原型和合成数据,
660	then train and validate inspection models	然后培训和验证检查模式
661	before production starts.	在开始生产之前。
662	Intelligence Center ex-agents,	情报中心前特工
663	built with NVIDIA's Factory Operations Blueprint,	与NVIDIA的工厂业务蓝图一起建造,
664	use orders, supplies and line setups	使用订单、用品和线路设置
665	to recommend production plans.	建议生产计划。
666	数字双人编译器使用 NVIDIA Omniverse 图书馆
667	to validate those plans with physics simulation	通过物理模拟验证这些计划
668	and real-time factory data	和实时工厂数据
669	before changes reach the factory floor.	在变化到达工厂地板之前。
670	With the Dassault Systems platform,	与达索系统平台,
671	an engineer needs to adapt a Volkswagen Golf Door design	工程师需要适应大众汽车高尔夫门的设计
672	for the Tiguan.	为提瓜人.
673	Aura 和 Leo 使用 NVIDIA 奈马特龙
674	to reason over engineering requirements.	在工程要求上讲道理
675	The harness coordinates the specialized sub-agents	带子协调专业的副代理
676	and preserves context across planning, design and simulation.	并保存整个规划、设计和模拟的背景。
677	Aura用Inovia来策划工作
678	around the vehicle's dimensions, weight limits	围绕车辆的尺寸、重量限度
679	and other engineering requirements.	和其他工程需求。
680	Leo Design Agent selects a compatible Golf Door design	狮子座设计代理选择兼容的高尔夫门设计
681	and uses Catia to reshape it for the Tiguan.	并用卡提亚重塑它为提瓜人。
682	Leo Simulation Agent uses Simulia,	里欧模拟特工用西穆利亚
683	由NVIDIA CUDA-X图书馆加速,
684	to test the variance, revealing a vibration problem.	以测试差异,显示振动问题。
685	The results return to the Design Agent for refinement.	结果回到设计代理公司进行完善.
686	The engineer reviews the refined designs	工程师审查精细设计
687	and selects the final one.	并选择最后一个。
688	Different goals, different expertise,	不同的目标,不同的专门知识,
689	the same building blocks,	同样的构件,
690	open models, CUDA-X accelerated tools	开放式模型, CUDA-X 加速工具
691	and the Open Shell Secure Runtime,	和开放的贝壳安全运行时间,
692	brought together with the NVIDIA Agent Toolkit.	与NVIDIA代理工具包合并。
693	Isn't that great?	是不是很棒吗?
694	Some people said that AI is here,	有人说AI在这里
695	it's gonna be the end of software.	这会是软件的终结。
696	It's gonna be end of tools company.	工具公司就要结束了
697	It's exactly the opposite.	则正相反.
698	AI is here, there'll be more users	AI在这里,会有更多的用户
699	of these tools that we created over time than ever before.	我们创造的这些工具
700	Every single tools company will now become an AI company.	现在每个工具公司都会成为AI公司.
701	They'll be an agents company.	他们会是代理公司。
702	Not only will they offer tools,	他们不但提供工具,
703	but they'll offer you specially optimized agents	但他们会提供你特别优化的特工
704	that are incredibly good at using their tools.	他们非常善于使用工具
705	Cadence and CrowdStrike and Dassault, IQVS	卡登斯和人群冲击和达索尔,IQVS
706	and just about every enterprise company out there	大约每个企业公司在那里
707	are all building agents	都是建筑经纪人
708	and we're just delighted to be working with them.	我们很高兴和他们合作
709	Building agents is not that hard.	建筑经纪人没那么难
710	It's got four, five components.	它有四个,五个组件。
711	It's got the model, the harness,	它有模型, 绳子,
712	both of them have closed and open versions,	他们俩都有封闭和开放的版本,
713	but the Open Shell Secure Runtime	但打开贝壳安全运行时间
714	and then most importantly, the knowledge and tools	最重要的是知识和工具
715	and those knowledge and tools belong to you.	那些知识和工具属于你
716	That's your expertise.	这就是你的专长
717	And so all of this is now possible to do	所以现在一切都可以了
718	and I'll show you more in just a second.	我马上就给你看更多
719	But before I do that, let's talk about physical AI.	但在我这样做之前, 让我们谈谈身体AI。
720	That's really the next wave of AI.	这真的是下一轮AI.
721	AI that understands the physical world	了解物理世界的AI
722	can reason about it because the physical world is diverse.	因为物理世界是多种多样的
723	It is complicated.	这很复杂。
724	It's got extremely long tails	尾巴很长
725	and to reason about it and use appropriate tools	并以此为理由,并使用适当的工具;
726	to interact with the physical world.	与物理世界互动。
727	Whenever people think about physical worlds,	每当人们想到物理世界
728	they think robotics.	他们认为机器人。
729	In fact, one of my favorite physical world applications	事实上,我最喜欢的物理世界应用之一
730	is the radio, cell phone, cell towers	是收音机、手机、手机塔
731	and the reason for that is this.	而原因就是这个。
732	It uses instead of cameras,	它用相机代替摄像头
733	instead of speakers and microphones,	而不是扬声器和麦克风,
734	it uses radio waves to sense the world	它用无线电波来感知世界
735	and it reasons about it, it processes information,	它的原因,它处理信息,
736	reasons about it and interacts with the world	与世界互动
737	using beamforming.	使用束状。
738	So in fact, this is an application, a technology	所以事实上,这是一个应用, 一个技术
739	that is just perfectly suited for AI.	这完全适合AI。
740	AI will revolutionize the way we do telecommunications.	AI将改变我们的电信方式。
741	It will enhance our spectrum efficiency tremendously	这将极大地提高我们的频谱效率
742	just as we've done so with ray tracing.	就像我们在追踪射线时所做的那样
743	All of the graphics that you've seen	所有你看到的图形
744	completely 100% ray traced.	完全100%的射线追踪。
745	The amount of computation we thought necessary in the past	我们过去认为必要的计算量
746	and now that we apply AI to is a million to one difference	而现在,我们应用AI 是一百万的区别
747	and so we're gonna bring to the telecommunications network	所以我们要带电信网络
748	an enormous step up in spectrum efficiency	* 频谱效率大幅提高
749	but beyond that, we're also gonna revolutionize	但除此之外,我们还要去革命
750	all of these cell towers.	所有这些细胞塔。
751	There's 100,000 sites around the world,	全世界有十万个网站
752	around the Western world,	在西方世界,
753	about 100 gigawatts of distributed data centers	分布式数据中心约100千兆瓦
754	that will in the future become	未来将变成
755	a distributed AI infrastructure.	a 分布式人工智能基础设施。
756	We're delighted to partner with many companies	我们很高兴和很多公司合作
757	here in Europe and just this last week,	在这里,在欧洲, 就在上周,
758	after working on this for nearly eight years,	经过近八年的努力
759	NVIDIA made the first commercial phone call	NVIDIA打了第一个商业电话
760	与诺基亚的T-Mobile网络
761	and so now CUDA makes phone calls, the traditional way.	因此,现在CUDA 打电话,传统的方式。
762	Then once it can do that,	一旦它能做到这一点,
763	then we can layer AI on top of it	然后我们就可以把人工智能分层了
764	and so the revolution is going to begin.	革命就要开始了
765	This is really exciting opportunity for us.	这对我们来说真是令人兴奋的机会。
766	The other physical AI application	另一个物理AI应用程序
767	that I'm super excited about	我很兴奋
768	is essentially robotic radiotherapy.	本质上是机器人放射疗法。
769	The tumor is changing in shape,	肿瘤在变形,
770	the physical body is changing while you're in therapy.	身体正在改变当你在治疗。
771	The robot, this adaptive radiotherapy system	机器人,这个适应性放射治疗系统
772	has to perceive the changing environment,	必须了解不断变化的环境,
773	understand where the tumor is,	了解肿瘤在哪里,
774	segment it just perfectly	片段完美无缺
775	and direct, precisely direct the radiotherapy accordingly.	并直接,精确地指导放射治疗
776	Essentially robotic, that's right, go ahead.	基本上,机器人,没错,继续。
777	This is a form of robotic surgery.	这是机器人手术的一种形式。
778	Now of course, the highest volume,	当然,最高的音量,
779	the most near term robotics application	最近期的机器人应用
780	or physical AI application that everybody recognized	或每个人都承认的物理AI应用
781	has been something we've been working on	我们一直在研究
782	for over a decade and now we recognize	十多年了,现在我们承认
783	that every single car company will want to make their cars	每个汽车公司都想制造他们的汽车
784	part of the ride hailing fleet.	部分的车驾冰雹舰队。
785	Today we have ride hailing and it's part of Uber	今天我们骑着冰雹,它是Uber的一部分
786	and here in Europe we have Bolt	在欧洲我们有博尔特
787	我们有Grab和Lyft等等
788	and all these different ride hailing services.	和所有这些不同的车欢呼服务。
789	In the future there'll be robo taxis	以后会有计程车
790	that will be connected to those ride hailing services	与那些欢呼服务相连
791	as well but it won't be from just one brand	也好,但不只是一个品牌
792	and so we recognize that building these systems	因此,我们认识到建立这些系统
793	are incredibly expensive,	非常昂贵,
794	particularly the software that goes on top of it.	特别是上面的软件
795	We decided to create one standard architecture,	我们决定建立一个标准建筑,
796	14 different cameras, nine radar systems,	14个不同的摄像机,9个雷达系统,
797	three lidars, umpteen ultrasonics,	三里达,超音速
798	this cocoon of a shell of sensors	这只探测器的茧
799	由双NVIDIA GPU系统供电
800	for the AV computing system.	为AV计算系统.
801	We turned this into an industry standard	我们把这个变成了行业标准
802	and now it's being designed into one car fleet	现在它被设计成一个车队
803	after another.	后别.
804	On top of it are AV software companies	上面还有AV软件公司
805	and we are in fact one of them as well	其实我们也是其中之一
806	to build an autonomous robo taxi capability	建设自动驾驶出租车能力;
807	which then is connected into the ride hailing services.	然后连接到冰雹服务中心
808	This ecosystem is being built out.	这个生态系统正在建设中。
809	The world travels about a trillion miles per year	世界每年旅行一万亿英里
810	around the world.	环绕世界.
811	Many of those miles in the future,	未来许多英里
812	you know a dollar or two per mile	你知道每英里一两块钱
813	are going to be on robo taxis.	将会是 robo出租车。
814	So this is one giant opportunity	因此,这是一个巨大的机会
815	but it's incredibly fragmented	但它令人难以置信的破碎
816	and so creating an industry standard, Hyperion,	于是创造了一个行业标准,Hyperion,
817	and working with an entire ecosystem,	与整个生态系统合作,
818	we could create essentially a giant virtual fleet	我们基本上可以建立一个巨大的虚拟舰队
819	that everybody can connect into.	每个人可以连接。
820	Today, I'm super proud to announce	今天,我非常骄傲地宣布
821	that four more brands, four more cars are going to,	4个品牌,4个汽车要,
822	lines are going to come in.	线路会进来的
823	The VW van, I just can't imagine anything more iconic.	越战车,我无法想象任何更具有标志性的东西。
824	The JLR brand new car, the Volvo fleet	JLR新车沃尔沃车队
825	as well as Mercedes next generation	还有下一代梅赛德斯
826	也会是Hyperion兼容的
827	and be part of this network,	并成为这个网络的一部分,
828	this ride hailing network, robo taxi fleet.	这车的冰雹网络, robo出租车车队。
829	欢迎来到Hyperion
830	The next major industry, the next major industry,	下一个大工业,下一个大工业,
831	the next great industry and this one is likely	下一个伟大的产业和这个有可能
832	to be extremely large is humanoid robotics.	极大的就是人造机器人。
833	There's a reason why robotics is hard to do	机器人很难做到是有原因的
834	because all the different work is rather diverse	因为所有不同的作品都相当不同
835	and so building a special configuration	因此,建立一个特殊配置
836	for each one of the diverse tasks	每个不同的任务
837	segments the robotics industry	机器人工业
838	into all of these different configurations	进入所有这些不同的配置
839	which prevents the technology flywheel to happen.	防止技术飞轮发生
840	The reason why the cell phone has advanced so quickly,	手机这么快推进的原因
841	the reason why personal computers have advanced so quickly,	个人电脑如此快速发展的原因
842	NVIDIA的NVLink 72如此快速推进的原因
843	is because those standardized configurations	这是因为这些标准化配置
844	allows many customers to use it,	允许很多顾客使用它,
845	the volume to therefore be derived,	由此而得之量,
846	that volume drives the technology flywheel	那个音量驱动着技术飞轮
847	and it gets faster, better, incredibly fast.	越快越好越快越好
848	And so we need a robotics configuration	所以我们需要机器人配置
849	that is as standard as possible.	这是尽可能标准。
850	Now human robotics obviously has the ability,	现在人类机器人显然有能力
851	it probably isn't perfect for everything	可能不是完美无缺
852	but it's likely to work for many things.	但它可能为很多事情工作。
853	As a result, this is likely going to be	因此,这可能是
854	the largest volume robotics configuration.	最大的量级机器人配置。
855	Now if we could just create the AI	现在,如果我们可以创建AI
856	that allows it to learn very quickly,	让它能很快学习
857	then for small medium businesses	中小型企业
858	who don't have an entire robotics team	没有一个机器人团队
859	to teach and program the robot	教授和编程机器人
860	can now show the robot what needs to be done.	现在可以告诉机器人需要做什么。
861	The robot because it has general capabilities	机器人因为拥有一般能力
862	and general intelligence	和一般情报
863	would be able to learn and adapt quickly.	能够迅速学习和适应。
864	And so this is likely to be the next major platform,	所以这很可能是下一个主要平台,
865	technology platform.	科技平台.
866	We're working with so many different companies,	我们和很多不同的公司合作
867	you will recognize many of them here in Europe.	在欧洲你会认出他们中的很多人
868	This is your great opportunity.	这是你的大好机会
869	No country, no region has dedicated itself	没有国家,没有地区有奉献
870	to making megatronics, electronics, mechanical systems	制造巨星、电子、机械系统
871	at the level of precision and craft as the European region.	作为欧洲区域的精确度和工艺度。
872	So many companies here are incredibly good at manufacturing.	这里有很多公司都非常擅长制造。
873	This is going to be the next high volume industry.	这将是下一个高额工业。
874	And you could see there's so many companies	你可以看到有很多公司
875	that are now part of it	现在成为它的一部分
876	and we're delighted to be partnered with you.	我们很高兴能跟你合作
877	Now in order to create a physical AI system,	现在为了建立物理AI系统,
878	you need three computers.	你需要三台电脑
879	The first computer is the one we've been talking about	第一台电脑是我们一直在谈论的
880	and everybody knows quite well	人人都很了解这个
881	which is training the AI model.	这是训练AI模式。
882	The second computer is where we have the AI	第二台电脑就是我们有AI的地方
883	learn how to do its skills.	学会如何发挥自己的技能。
884	You can't just have the AI,	你不能只是拥有AI,
885	program the AI and have the AI come out	编程AI 让AI出来
886	into the physical world	进入物理世界
887	and try to do whatever it wants to do.	试着做它想做的事
888	You do it inside a digital twin.	你在数码双胞胎内部做
889	We create a digital twin called Omniverse	我们创造了一个名为Omniverse的数字双胞胎
890	and you've seen us talk about that repeatedly	你见过我们反复谈论
891	and in fact just now,	事实上就在刚才
892	when you saw all the CUDA-X libraries,	当你看到所有的 CUDA -X 库,
893	they're all being done inside a digital twin world.	它们都在数码双胞胎世界里完成
894	That digital twin world is called Omniverse.	那个数码双胞胎世界叫做Omniverse。
895	Omniverse是一个连接的库
896	into all kinds of platforms and tools,	进入各种平台和工具,
897	Siemens and Dassault and Cadence and so on and so forth.	西门子和达索尔,卡登斯等.
898	And so you could see these digital twin systems	这样你就能看到这些数码双子系统
899	being created.	正在创建。
900	The second computer.	第二台电脑
901	The third computer is the robotics computer.	第三台计算机是机器人计算机.
902	That computer integrates in the case of	计算机集成到
903	the self-driving car, the robotaxi,	自驾车机器人轴
904	我们称它为Hyperion。
905	Hyperion由电脑本身组成,
906	the functional safety system,	功能安全系统,
907	the active safety system,	积极的安全系统,
908	an operating system designed for physical AI safety.	用于物理AI安全的操作系统。
909	Well, we need to do exactly the same for robotic systems.	我们需要对机器人系统做同样的事情。
910	So that every single time you start to develop research	所以每次你开始研究
911	and start to work on your own robotic systems,	开始研究你自己的机器人系统
912	you don't have to start from scratch	你不必从头开始
913	to cobble up all of the software and computing systems	将所有软件和计算系统拼接起来
914	and the sensors and all the mechanical systems.	传感器和所有机械系统
915	That alone would take several years.	光这样就需要几年时间。
916	And so we decided to build a reference robot.	所以我们决定建立一个参考机器人。
917	这是NVIDIA AGILE一号搭档
918	与Agile机器人学。
919	This robot is going,	这机器人要飞了
920	this reference robot will be available	此参考机器人将可用
921	in the first quarter of next year	下年第一季度
922	and it will come pre-assembled with computer inside,	里面有电脑,
923	software integrated, and it should be able to	软件集成,它应当能够
924	perform quite well right out of the bat.	表现很好右出蝙蝠。
925	And so this is a great starting point	因此,这是一个伟大的起点
926	and it'll be the flywheel by which we can accelerate	它将会是飞轮我们可以加速
927	development of robotics.	机器人的发展。
928	Well, let me start.	好吧,让我开始。
929	Let me end today's speech from where we started.	请允许我从我们开始的地方结束今天的发言。
930	NVIDIA起源于PC革命.
931	It was in 1993 and the PC revolution	是在1993年和PC革命期间
932	was just about to start.	刚刚开始。
933	Remember, Windows 95 kicked it into Turbo Charge.	记住,Windows 95把它踢进涡轮冲锋机
934	We came since then, invented the GPU,	从那时起,我们发明了GPU
935	programmable shading, real-time ray tracing.	可编程阴影,实时射线跟踪.
936	We added CUDA to the computing model	我们在计算模型中加入了CUDA
937	and helped ignite the modern AI revolution.	并且帮助点燃了现代AI革命.
938	And so four years ago, we started to think,	所以四年前我们开始思考
939	what can we do to reimagine the personal computer	我们能做些什么来重新想象个人电脑
940	for this new era?	对于这个新时代?
941	Where the personal computer is not just your tool,	如果个人电脑不仅仅是你的工具
942	your personal computer tool,	您的个人电脑工具,
943	but your personal agent or your personal assistant.	但你的私人经纪人或私人助理
944	And the way you interact with this computer	你和这台电脑互动的方式
945	would be essentially a companion,	在那日,他确是一个伴侣,
946	somebody who helps you do work.	帮助你工作的人
947	And so this computer needs to have exactly the five things	所以,这台电脑需要准确的五件事
948	that I mentioned today.	我今天提到过
949	It needs to have model.	它需要模型。
950	It could be running locally	可能在本地运行
951	or it could be running in the cloud.	或者它可能在云中运行。
952	It needs a harness.	它需要一把吊带。
953	It needs a secure runtime, open shell,	它需要一个安全的运行时间,打开的壳,
954	and it needs access because it's available on your PC.	它需要访问,因为它在你的电脑上。
955	It needs access to knowledge and tools.	它需要获得知识和工具。
956	And the good news is because NVIDIA has been accelerating	好消息是NVIDIA已经加速了
957	tools for so long.	工具这么久。
958	If we could just create a computer	如果我们能创建一台电脑
959	可以加速 OpenGL 和 DirectX
960	and everything CUDA,	和一切库达,
961	NVIDIA做过的一切
962	would run on this computer.	会用这台电脑运行
963	And so this is it.	这就是它。
964	After four years of partnership with Microsoft,	在和微软合作四年后,
965	我们已经创建了RTX Spark.
966	And this is a brand new line of computers.	这是全新的电脑线路。
967	Literally thousands of years,	数千年,
968	thousands of engineering years went into this.	几千年的工程年进入了这个。
969	We've been secretly working on it for four years,	我们秘密研究了四年
970	created a chip called the N1X	创建一个名为 N1X 的芯片
971	and then we just recently announced it	然后我们刚刚宣布
972	with the entire PC industry joining forces.	与整个PC行业联手.
973	We created desktops, laptops, and workstations.	我们创建了台式计算机、笔记本电脑和工作站。
974	This is one petaflops.	这是一只宠物
975	One petaflops, just to put that in perspective,	一只宠物,只是说到这一点,
976	in the year 2016,	2016年时,
977	I delivered one petaflops	我送了一只燕窝
978	and 128 gigabytes of memory.	还有128千兆字节的记忆
979	One petaflops and 128 gigabytes	一只燕窝和128千兆字节
980	to a small company called OpenAI.	给一家叫OpenAI的小公司
981	This is one petaflops and 128 gigabytes.	这是一只燕窝和128千兆字节。
982	The only difference, of course,	当然,唯一的区别
983	is that I can hold this like this	我可以像这样拿着它
984	and DGX One took two people to carry upstairs.	和总干事一个带两个人上楼
985	这是全新的 RTX Spark PC。
986	And if you would like to have a companion workstation	如果你想有一个伴工站
987	that runs Katia and everything,	经营卡蒂亚和一切
988	自动桌面和Adobe
989	and Cadence and Synopsys,	卡登斯和赛诺斯人,
990	you can always get yourself a DGX station	你总能给自己找个DGX站
991	that runs Windows.	运行窗口。
992	And this is 20 petaflops	这是20个宠物
993	and 754 gigabytes of memory.	还有754千兆字节的记忆
994	And so incredible.	如此不可思议.
995	And of course, there's also the desktop version.	当然,还有桌面版本.
996	All of these are now available	现在所有这一切都可以使用。
997	我认为DGX站与RTX Spark,
998	使用 DGX Spark, DGX 站使用 Windows
999	will be available at the end of this year.	将于今年年底提供。
1000	Now the question is,	现在的问题是,
1001	so far I just talked about the five components of an agent	到目前为止,我刚刚谈到一个特工的五个组成部分
1002	and now the question is, how do you build it?	现在的问题是,你是如何建造的?
1003	How do you build an agent that runs,	如何建立一个代理运行,
1004	let's say on one of those computers	说到其中一台电脑
1005	or runs in a cloud or runs on-prem somewhere?	或者在云中奔跑或者在某个地方奔跑?
1006	Well, the answer is it's really easy	答案是,这很简单
1007	and you do it this way.	你这样做。
1008	The way you build an agent is you ask an agent	你建立特工的方式就是问一个特工
1009	to build it for you.	为你建造它。
1010	So this, this is the prompt	所以,这个是提示
1011	that is going to run the next application.	这将运行下一个应用程序。
1012	And what you're about to see is this.	你将要看到的就是这个。
1013	You're about to see a robotics application.	你即将看到机器人的应用。
1014	And while you're watching this demo,	当你在看这个演示时
1015	just remember there's an agent running	记得有个特工在跑
1016	on this personal computer.	在个人电脑上
1017	This demo runs across three computers.	此演示横跨三台电脑.
1018	The first computer is to train.	第一台电脑是用来训练的.
1019	The second is a digital twin.	第二种是数字双胞胎.
1020	And the third is to put that AI into the computer itself.	第三是把人工智能放进电脑里
1021	Okay, so two very, two ideas.	好吧,所以两个非常,两个想法。
1022	The first idea is agents,	第一个想法是特工
1023	how do you build it and run it	如何建造和运行它
1024	and what you can use it to do.	还有你能用它做什么
1025	And the second is physical AI computing model,	第二个是物理AI计算模型,
1026	which consists of three computers.	由三台计算机组成。
1027	Let's take a look.	让我们来看看。
1028	见到Microduck,拥抱面部机器人。
1029	让我们教它看和走与Nvidia RTX Spark。
1030	I use Hermes as my agent harness,	我以赫尔墨斯为经纪人
1031	Nematron和OpenAI的GPT-6 Astra作为我的模特儿
1032	以及CUDA-X加速工具、FreeCAD、Blender和IsaacSim,
1033	Newton Physics Engine and IsaacLab.	牛顿物理引擎和艾萨克拉布.
1034	我会运行所有我的RTX Spark,
1035	except when I need to burst larger workloads to cloud.	除非我需要把更大的工作量推向云层
1036	First, I describe the course	首先,我描述课程
1037	and what my Microduck needs to learn.	还有我的Microduck需要学的东西
1038	On RTX Spark, Hermes and Nemotron plan the course,	在RTX Spark上,赫尔墨斯和尼莫特龙计划了课程,
1039	listing every object required.	列出每个需要的对象。
1040	Astra takes it from there,	阿斯特拉从那里拿走了
1041	使用 FreeCAD 模拟 3D 对象,
1042	Blender 用于应用材料
1043	和 Omniverse SimReady 库以添加物理属性。
1044	In IsaacSim, Astra composes the scene.	在艾萨克辛(IsaacSim)中,阿斯特拉(Astra)为场景作曲.
1045	Astra钻机 Microduck的关节、引爆器和链接
1046	and simulates how Microduck moves with Newton Physics Engine.	并模拟Microduck如何用牛顿物理引擎移动.
1047	Now, training begins.	现在,训练开始了。
1048	In the cloud, thousands of Microducks train in parallel in IsaacLab.	在云中,以艾萨克拉布的数千辆微电车平行运行.
1049	Every fall, a lesson.	每年秋天,一个教训。
1050	Improving the policy every time.	每次完善政策.
1051	模拟Microduck的摄像机,
1052	teaching it to see.	教其观心.
1053	With perception and physics working together,	有了感知和物理合作,
1054	my Microduck can avoid obstacles, navigate tight spaces	我的Microduck可以避开障碍, 导航紧凑的空间
1055	and adapt to its surroundings.	并适应其周围环境。
1056	我把保险单下载到RTX Spark
1057	and use Nemotron and IsaacSim to evaluate my Microduck	利用尼莫特隆和艾萨克辛来评价我的Microduck
1058	with multi-robot orchestration.	和多机器人的管弦乐
1059	I deploy the validated policy to the real duck	我将经过验证的保单用在了真正的鸭子上
1060	and he's on his way.	他马上就到
1061	NVIDIA RTX Spark. (原始内容存档于2018-07-29).
1062	Build and run AI agents and their accelerated tools	构建和运行人工智能代理及其加速工具
1063	right on your laptop.	在你的笔记本电脑上。
1064	Hey, guys.	嘿,伙计们。
1065	Oh, yeah?	是吗?
1066	Okay, come on.	好吧,来吧。
1067	Here, come over here.	过来,过来
1068	Show people you can walk.	向人们展示你可以走路。
1069	You have a camera.	你有一个相机。
1070	Oh, yeah?	是吗?
1071	Don't trip.	不要绊倒。
1072	Can you guys sit down?	你们能坐下吗?
1073	Okay, stand up.	好吧,站起来。
1074	Take a nap.	睡个觉。
1075	You need to take a nap.	你得睡个觉
1076	Okay, stand up.	好吧,站起来。
1077	Did you know that you were created in here?	你知道你是在这里创造的吗?
1078	你是在RTX Spark上创建的.
1079	I know.	我知道,我记得
1080	And trained you in the cloud	在云中训练你
1081	on a whole bunch of NVIDIA supercomputers.	在一堆NVIDIA超级计算机上
1082	And you learned your skills right here on RTX Spark	你在RTX Spark上学到了自己的技能
1083	in a digital twin, the Omniverse digital twin.	在数字双胞胎中,Omniverse数字双胞胎.
1084	Isn't that incredible?	这不是不可思议吗?
1085	I know.	我知道,我记得
1086	You guys are so cute.	你们真可爱
1087	And your parents are called Pollen.	而父母名为波伦.
1088	P -O -L -L
1089	Uh-huh.	对
1090	Pollen Robotics.	波伦机器人
1091	And apparently, everybody loves you guys so much.	显然,每个人都爱你们这么多。
1092	You're sold out.	你卖光了。
1093	I know, so if you go to the Pollen website, everybody,	我知道,所以如果你去波伦网站,每个人,
1094	everybody watching this,	大家看这个
1095	you know, there's going to be like 20 million viewers,	会有两千万观众
1096	everybody watching this, go to the Pollen website	大家看这个,去波伦的网站
1097	and place an order.	并发布命令。
1098	I think this startup company is like 10 people.	我觉得这家创业公司像10个人
1099	They might blow up.	他们可能会爆炸。
1100	It's going to be so exciting.	一定会很刺激的
1101	Everybody should have one.	每个人都应该有一个。
1102	You guys are so cute.	你们真可爱
1103	And so we'll just have an RTX Spark	所以,我们只需要一个RTX Spark
1104	and we'll teach you new skills.	我们会教你新的技能
1105	Isn't that right?	是吗?
1106	Oh, look at that.	哦,看看那个。
1107	That's incredible.	不可思议.
1108	Can you do a somersault?	你能做点什么吗?
1109	That's so good.	赞曰.
1110	Wow.	珃
1111	Okay.	摆
1112	Well, guess what?	你猜怎么着?
1113	Okay, so today, what did we talk about?	好吧,那么今天,我们谈了什么?
1114	We talked about Europe building its AI economy.	我们谈到了欧洲建设其AI经济.
1115	And it starts with infrastructure.	它从基础设施开始。
1116	And we announced a brand new,	我们宣布了一个全新的,
1117	brand new system called DGX Pod.	全新的系统叫做DGX Pod.
1118	It makes it easy for you to build AI factories.	这让你很容易建造AI工厂。
1119	And these AI factories don't just run models anymore.	而这些AI工厂不再只是经营模特.
1120	Pay attention.	专心点
1121	Stay focused.	专心点
1122	Come here, come here.	过来,过来
1123	Come over here.	过来
1124	Stay focused.	专心点
1125	And these AI factories don't just run,	这些人工智能工厂不只是运行,
1126	there you go, sit down, yeah.	给你,坐下,是的。
1127	They don't just run AI models anymore.	他们不只是运行AI模型了。
1128	They run entire agents.	他们管理着整个特工
1129	And these agents are consistent of five things.	这些特工有五样东西
1130	There's the large language model that does the perception,	有大型语言模型做感知,
1131	reasoning, planning, and acting.	推理、规划和行动。
1132	And it runs in conjunction with a harness	它和绳子一起运行
1133	that turns it essentially exoskeleton.	基本上,它变成了外骨骼。
1134	And once that harness sits inside,	一旦绳子坐在里面,
1135	runs inside a secure runtime we call open shell,	运行在一个安全的运行时间我们称打开壳,
1136	which has just so many people that have supported it.	它有这么多人支持它。
1137	And you give it access to knowledge and tools.	你让它获得知识和工具。
1138	And this agent runs on top of these computers,	这个特工在电脑上运行
1139	but we're also extending into the next generation,	但我们也延续到下一代,
1140	which is where you guys come in, physical AI.	这就是你们进来的地方,身体AI。
1141	So that's it.	故曰此义.
1142	That's what we talked about.	这就是我们谈论的。
1143	Everybody can build agents.	每个人都可以建立特工。
1144	Every company in Europe will become an AI company.	欧洲的每一个公司都会成为AI公司.
1145	There's no question everybody will become an AI company.	毫无疑问每个人都会成为AI公司
1146	And we're going to have a thriving AI economy here in Europe.	我们将在欧洲有一个繁荣的AI经济。
1147	So after 33 years, what an incredible journey.	所以33年后,多么不可思议的旅程。
1148	And we made a really cool video.	我们做了一个很酷的视频。
1149	You guys want to watch it together?	你们想一起看吗?
1150	Okay, come on.	好吧,来吧。
1151	Come over here.	过来
1152	Just come over here just a little bit.	只是过来一点点。
1153	We'll get out of the way.	师便出.
1154	Come on.	来吧。
1155	Top speed.	顶速.
1156	How the heck are we supposed to follow that?	我们该怎么跟他们走?
1157	Well, there is one thing that can make this crowd go even crazier.	有一件事能让人群更加疯狂
1158	What is it?	这是什么?
1159	You won't believe this!	你不会相信的!
1160	GTC has just gone full Eurovision!	GTC刚刚走全欧视!
1161	It begins.	它开始。
1162	Agents working by your side.	你身边的特工
1163	Europe's moment has arrived.	欧洲的时刻到了。
1164	But if you missed what we said at GTC,	但如果你错过我们在GTC说过的话
1165	we'll break it down for you.	我们会帮你破解的
1166	Germany!	德国!
1167	This is a story all about how the world's built.	这是一个关于世界如何建设的故事。
1168	First power plants let energy scale.	第一发电厂允许能源规模。
1169	Then came data centers.	然后是数据中心
1170	Now we've got mail.	现在我们有邮件了
1171	Now we'll be past turning coal into watts.	现在我们可以把煤变成瓦特了
1172	AI factories turn sand into thoughts.	AI工厂将沙子变成思想.
1173	Move over oil to S&P too.	将石油也移到 S&P 。
1174	There's something more investable.	还有更值得投资的
1175	我们的GPU。
1176	It works like an asset.	它的工作就像一个资产。
1177	Reducing green.	减少绿色。
1178	Looks like a chip.	看起来像芯片
1179	Acts like a cash machine.	表现得像个现金机器
1180	They make revenue while you relax.	你放松的时候,它们能赚钱
1181	The cash flows smoother than this sun.	现金流比太阳更顺畅
1182	Agents code and make decisions.	代理密码和决定。
1183	Give them freedom but control their missions.	给他们自由,但控制他们的任务
1184	Open shell on bearers.	打开背包
1185	Checking the list.	检查名单。
1186	No credentials.	无证书。
1187	Access dismissed.	禁止进入
1188	Sentry on blue field.	蓝田哨兵.
1189	Never clocks off.	永远不要关闭时钟。
1190	Protecting the sandbox.	护持沙盒.
1191	Like Hasselhoff.	就像哈塞尔霍夫
1192	We advance AI but limit risks.	我们推进AI,但限制风险。
1193	We're responsibly optimistic.	我们很乐观
1194	RTX Spark使特工安全而明亮.
1195	Our spot rates keep the price just right.	我们的现货价格保持了正确的价格。
1196	We do quantum rexes.	我们做量子量轴。
1197	Gaming all in one.	一起赌博
1198	Cause you can't spell fungible without fung.	因为你不能拼写可替换没有可替换。
1199	Now it's learning the rules of gravity friction.	现在它正在学习重力摩擦的规则.
1200	And using real tools.	并使用真正的工具。
1201	With Isaac Lab and Cosmos.	与艾萨克实验室和宇宙。
1202	They train on their own.	他们自己训练。
1203	It's like college for robots.	这就像大学机器人。
1204	Without student loans.	没有学生贷款。
1205	DSX pocks同步给出AI.
1206	Less power, more brains.	力量少,头脑多.
1207	Like LEGO's that thing.	就像LEGO就是这样
1208	QVS,Neymatron, 和库莫在这里。
1209	Throwing knowledge faster than a Berlin cashier.	比柏林的收银员还快
1210	Your full stack.	你满堆的
1211	It's all on its checklist.	全都在清单上
1212	AI's five layer cake.	AI的五层蛋糕.
1213	It eats it for breakfast.	它吃它作为早餐。
1214	Eurracle系统领先赛事.
1215	It loves open source.	爱开源.
1216	Just ask.	便问.
1217	Hug and faith.	拥抱与信仰.
1218	The continent is rising.	大陆正在崛起。
1219	And the best is yet to come.	而最上者未至.
1220	Come on Europe.	来吧欧洲。
1221	Let's sing as 1TC.	让我们唱为1TC。
1222	Welcome to GTC.	欢迎来到GTC.
1223	I want to thank all of you.	我想感谢你们所有人。
1224	All the sponsors.	所有赞助者.
1225	You can see all the names of companies that are here.	你可以看到这里所有公司的名字.
1226	All the partners we've been working with for 33 years.	我们合作了33年的所有合伙人
1227	Microduck.	微达克.
1228	My new friends.	我的新朋友们
1229	Let's work together to build Europe's AI economy	我们一起建设欧洲的AI经济
1230	and put AI to work.	并让AI工作。
1231	All right.	好吧,我们走。
1232	Have a great GTC.	有一个伟大的GTC。

该视频共有字幕 1240 条。解锁更多字幕为会员功能，请移动到价格

P3 · 【双语音+文稿】（完整版）2026 英伟达GTC柏林大会：英伟达创始人兼CEO黄仁勋发表主旨演讲！ p03 无字幕版 (P3)

【双语音+文稿】（完整版）2026 英伟达GTC柏林大会：英伟达创始人兼CEO黄仁勋发表主旨演讲！ p03 无字幕版 (P3)

发布于 2026-10-22 · 时长 1:33:26

字幕摘录

时间	英文	中文
0:09	This is how intelligence is made a new kind of factory	这就是智能制造的新型工厂
0:16	generator of tokens	符号生成器
0:19	building blocks of AI	A. 大赦国际的组成部分
0:25	Tokens have opened a new frontier	托肯斯开辟了新的疆域
0:29	Turning data into knowledge and knowledge into action	将数据转化为知识转化为行动
0:38	They decipher nature's code to reveal hidden cures	他们破译自然的密码揭示隐藏的解药
0:48	And mirror our cities	照亮我们的城市
0:57	And tirelessly navigate our world	不知疲倦地穿越我们的世界
1:01	To keep us moving	继续前进
1:06	Tokens learn from the way we move	托肯斯从我们的行动中学习

展开字幕全文（1232 条）

序号	英文	中文
1	This is how intelligence is made a new kind of factory	这就是智能制造的新型工厂
2	generator of tokens	符号生成器
3	building blocks of AI	A. 大赦国际的组成部分
4	Tokens have opened a new frontier	托肯斯开辟了新的疆域
5	Turning data into knowledge and knowledge into action	将数据转化为知识转化为行动
6	They decipher nature's code to reveal hidden cures	他们破译自然的密码揭示隐藏的解药
7	And mirror our cities	照亮我们的城市
8	And tirelessly navigate our world	不知疲倦地穿越我们的世界
9	To keep us moving	继续前进
10	Tokens learn from the way we move	托肯斯从我们的行动中学习
11	Deliver the care we need	提供我们需要的护理
12	And walk into danger	走进危险之中
13	To protect the ones we love	为了保护我们爱的人
14	They see what we cannot	他们看到我们所不能
15	So the steadiest hands never work alone	所以最稳健的手永远不能单独工作
16	Every idea Tokens spark	每一个想法托肯斯火花
17	Every success they share	他们分享的每一个成功
18	Puts what matters within reach	将重要的事物放在眼前
19	Tokens are helping us break new ground	托肯斯帮助我们开拓新天地
20	On a scale never seen before	在前所未有的规模上
21	So we can reach	这样我们就能到达
22	For the sky and beyond	为了天空和天空之外
23	Together we take the next great leap	我们一起走下一个大跃进
24	Into a bright new future	进入光明的新未来
25	Built for all mankind	为全人类而建
26	And here is where it all begins	在这里,一切都开始
27	欢迎来到舞台 NVIDIA创始人兼CEOJensen Wong.
28	Welcome to GTC.	欢迎来到GTC.
29	It's great to be here.	赞曰.
30	Before we start, let's give it up for our pre-game show hosts, smart and adorable, smart and adorable.	在开始前,让我们放弃我们的预赛节目主持人,聪明可爱,聪明可爱.
31	They spoke about AI infrastructure, sovereign AI, AI safety, and physical AI.	他们谈到了AI基础设施,主权AI,AI安全,以及实体AI.
32	They covered an enormous amount of ground this morning, and we are going to take it one step further.	今天上午,它们覆盖了大量地面,我们将进一步推进。
33	有趣的是,L'Oreal、Revolut、Spotify、Merck、Vodafone、Zara、保时捷、Nestle和汇丰银行有什么共同点?
34	Well, they're all European companies and they're all working with NVIDIA on AI.	他们都是欧洲公司他们都和NVIDIA合作研究AI
35	Everybody, everybody here is all activated on the AI ecosystem and working towards the AI economy.	各位,这里的每个人都在人工智能生态系统上启动并致力于人工智能经济
36	We've been working here in Europe now for 33 years.	我们已经在欧洲工作了33年
37	You probably don't realize that NV1, NVIDIA's first chip, NV2, and NV3, which became Revot 128,	你可能没有意识到NV1,NVIDIA的第一个芯片,NV2和NV3,它变成了Revot 128,
38	the chip that saved our company really, and launched our company, were all built in Europe.	真正拯救我们公司的芯片, 并推出我们的公司, 都是在欧洲建造的。
39	我们的第一个搭档是SGS Thompson
40	We manufactured a chip in Kroll, France, assembled it in Malta, and shipped it around the world.	我们在法国的克罗尔制造了芯片, 在马耳他组装,并运至世界各地。
41	Our first largest customer, the largest customer of GeForce Days in the beginning, was a company called MediaMark.	我们的第一个最大的客户,GeForce Days在一开始最大的客户是一家名为MediaMark的公司.
42	Do you know them?	你认识他们吗?
43	MediaMark had a spectacular idea about building PCs, as you remember, and they were our largest customer.	MediaMark对建造个人电脑有一个惊人的想法, 你还记得, 他们是我们最大的客户。
44	It was Elsa, a German company that brought NVIDIA into the workstation industry in the year 2000.	是德国公司Elsa在2000年将NVIDIA带入工作站产业.
45	Now, NVIDIA is in every single workstation all over Europe designing cars,	现在,NVIDIA在每一个工作站在整个欧洲设计汽车,
46	everything from styling, design, engineering, simulation, all the way through to even inside the car,	从造型、设计、工程、模拟到车内
47	which a European company brought us into, Audi, in the year 2005.	是一家欧洲公司在2005年带我们来的奥迪
48	Then in year 2020, we started working on autonomous vehicles with Mercedes.	之后在2020年,我们开始与梅赛德斯一起从事自主车辆的工作.
49	As you can see, NVIDIA's life journey, the entire 33 years, I've spent a lot of time in Germany,	如你所见,NVIDIA的人生历程, 整个33年,我花了很多时间在德国,
50	so I'm delighted to be back.	所以我很高兴能回来
51	Thank you.	谢谢
52	I'm delighted to do GTC here.	我很高兴在这里做GTC。
53	This year, this last year, we started the new journey to building a new industry here in Europe,	今年,去年,我们开始了在欧洲建设新产业的新征程,
54	the beginnings of the AI economy.	AI经济的开始.
55	In the last year alone, we've invested nearly $10 billion in Europe across startups,	仅在去年我们在欧洲投资了将近100亿美元跨创业企业
56	big partnerships, small partnerships, all over Europe.	大型伙伴关系,小型伙伴关系,遍布欧洲。
57	This last week alone, as I journeyed from France and UK to Sweden and now here in Berlin,	仅上周,当我从法国和英国前往瑞典,现在在柏林,
58	we announced the investments of 14 startup companies.	我们宣布了14家创业公司的投资
59	Overall, since the beginning, we now have 12,000-plus partnerships across Europe,	总体而言,自一开始,我们在欧洲各地建立了12,000多个伙伴关系。
60	and we're all working towards one goal now.	我们都在努力争取一个目标
61	After decades, after decades of advancement in the computer industry,	经过几十年的计算机工业的进步,
62	we are now at the beginnings of a new industrial revolution.	我们现在正处于新的工业革命的开始阶段。
63	This image was generated by a computer.	这个图像是由计算机生成的.
64	It was prompted by me.	是因为我
65	Nice, simple prompt about AI powering a new industrial revolution generated this.	人工智能发动新的工业革命的简单而明智的提示导致了这个结果
66	I gave it a few partners that I wanted to highlight, companies that we're working with,	我给了它几个合伙人我想强调, 我们合作的公司,
67	and this is an extraordinary thing.	这是一件不寻常的事。
68	In the mind of the AI, this is what it means to build a new industry,	在AI的心目中,这就是建设新产业的意义,
69	generation of tokens, the production of intelligence,	创造象征创造智慧
70	which integrates into every single industry powering a new industrial revolution.	它融入每一个工业推动新的工业革命。
71	A new industrial revolution happens when the new technology affects how we build something	新的工业革命会发生当新技术影响我们如何建造东西
72	and what we can build, how we build something and what we can build.	以及我们能够建造的东西,我们如何建造东西和我们所能建造的东西。
73	I think in every single way, artificial intelligence will affect everything we build	我认为,在每一种方式, 人工智能会影响我们建立的一切
74	and in fact enable us to build new things.	并且事实上使我们能够创造新的东西。
75	In this way, because everyone will use it, every company will be powered by it,	这样,因为每个人都会使用它, 每个公司都会得到它的力量,
76	because it's digital intelligence, the foundation of everything.	因为它是数字智能, 一切的基础。
77	Every single person will use it, every company will be powered by it,	每个人都会用它每一个公司都会用它
78	every industry will be transformed, and new ones will be created.	每一个产业都会被改造,新的产业也会被创造出来.
79	In this way, like the electricity, it is a new industrial revolution,	这样,就像电力一样,这是一场新的工业革命,
80	and we're at the very beginnings of that.	我们正处在一个开始。
81	The question is, what is AI?	问题是,AI是什么?
82	And this is a big deal.	这是件大事。
83	If you want to be part of creating this future, understanding what is AI is vitally important.	如果你想参与创造这个未来,了解AI是什么至关重要。
84	And this is really a simple framework.	这是一个很简单的框架。
85	We were early to observe that in fact AI is a manufacturing industry.	我们很早就注意到,事实上AI是一个制造业。
86	AI is not just software technology.	AI不仅仅是软件技术.
87	It is software technology, but not just.	这是软件技术,但不只是。
88	AI, of course, runs on a computer, but the way that that computer operates is fundamentally different.	当然,AI运行在一台计算机上,但该计算机的操作方式根本不同.
89	Rather than a tool that's used by us, it is now part of a machinery and infrastructure,	现在它不是我们使用的工具而是机器和基础设施的一部分
90	factories that are producing intelligence that are integrated into every single industry and application.	生产情报的工厂已经融入每一个行业和应用
91	A simple model is to think of AI as a five-layer cake.	一个简单的模型是把AI看作是一个五层的蛋糕.
92	Energy powers advanced computers that are part of an AI factory that generates tokens,	能源动力高级计算机属于AI工厂的一部分生成符号,
93	digital intelligence from the models that run on it that are integrated into every application.	来自运行在它上的模型中的数字智能被整合到每一个应用程序中.
94	A five-layer cake.	五层的蛋糕
95	To be part of this AI economy, you don't want to just be consuming the AI.	为了参与AI经济,你并不想仅仅消耗AI.
96	You want to be producing the AI.	你想制作人工智能。
97	We could do all of that here in Europe.	我们可以在欧洲做这一切。
98	And in fact, we've already started.	事实上,我们已经开始了。
99	Europe is building its AI economy and its five-layer cake represented here.	欧洲正在建设其AI经济及其代表在这里的五层蛋糕.
100	We've got great partnerships, and when you look across this five-layer cake,	我们有很好的合作关系, 当你看到这个五层蛋糕,
101	you can recognize that in fact it is a new industry because there's growth at every single layer.	你可以认识到,事实上这是一个新的产业, 因为每一个层次都有增长。
102	We're creating jobs, creating growth at an extraordinary rate.	我们正在创造工作机会, 以非凡的速度创造增长。
103	Companies here in Europe, Schneider, Siemens Electric, Siemens Energy, doing incredibly well around the world.	欧洲的公司,施耐德公司,西门子电气公司,西门子能源公司, 在世界各地都表现非常好。
104	We're the computing layer.	我们是计算层。
105	We integrate into AI factories that are now flourishing all over Europe.	我们融入了整个欧洲的人工智能工厂。
106	Startup companies like Nebia is doing incredibly well.	像Nebia这样的创业公司表现的非常好
107	A new company called Firebird, which is setting up a new cloud in the Central Asia region.	一家名为火鸟的新公司,正在中亚地区建立新云.
108	And even Cassava in Africa.	甚至非洲的木薯。
109	Every single telco in the future, and I predict this, every single telco in the future,	未来每一个Telco, 我预测,未来每一个Telco,
110	with a hundred thousand sites representing some hundred gigawatts of distributed telco base stations and data centers,	拥有10万个站点,代表着大约100千兆瓦的分布式电视机基站和数据中心,
111	will in the future evolve into a distributed AI network.	未来将发展成分布式AI网络.
112	So the AI infrastructure layer is being built out, which is then above that, the AI models.	因此AI基础设施层正在建设中,然后就超过了这个,AI模型.
113	It's really important to recognize that long before the AI models emerge, the infrastructure has to exist.	真正重要的是要认识到,早在AI模型出现之前,基础设施就必须存在.
114	Just as before you can have an internet, you have to have the networking.	和之前的互联网一样, 你必须有网络。
115	Just as before you could really enjoy the capabilities of electricity, you have to have electrified houses, electrified power plants.	就像以前你真正能够享受电力能力一样, 你必须拥有电气化的房屋,电气化的发电厂。
116	And so the models are the next layer.	所以模型是下一层。
117	It's also interesting to observe that the model layer is not just language models.	同样有趣的是观察到模型层不仅仅是语言模型.
118	It represents language, biology, chemicals, physics, robotic actions, animation, and it represents models of all sizes.	它代表语言,生物学,化学,物理,机器人动作,动画,它代表各种大小的模型.
119	These different industries are going to be powered by different modalities of AI models,	这些不同的产业将会通过不同模式的AI模式获得动力,
120	which then integrate into the AI natives, the startups.	然后融入人工智能本土,即初创企业。
121	And there are so many here in Europe.	欧洲有很多人
122	So the Europe's AI economy is starting, and it's really accelerating.	因此,欧洲的AI经济开始,它真的在加速.
123	It's great to see.	赞曰.
124	At the foundation of it, where we contribute, most people think is GPUs.	在它的基础,我们在那里贡献,大多数人认为是GPU.
125	It is true.	这是真的。
126	We started in 1993 building a graphics chip, which emerged into the GPU, the programmable shading GPU.	我们从1993年开始建造图形芯片,它出现在GPU,可编程阴影GPU中.
127	Over time, we added CUDA acceleration to that, reinvented computing as we know it, which then brought forth artificial intelligence.	随着时间的推移,我们添加了CUDA加速度, 重新发明了我们所知的计算, 从而带来了人工智能。
128	But NVIDIA today is really a full stack, end-to-end, complete AI factory platform.	但今天的NVIDIA其实是一个完整的堆栈,端到端,完整的AI工厂平台.
129	The breadth of the technology of our company over the years has really expanded.	这些年来,我们公司技术的广度确实得到了扩大。
130	Everything from GPU computing, Vera Rubin is our latest generation, to the NVLink switches that connect them into one giant GPU.	从GPU计算,Vera Rubin是我们最新的一代, 到连接到一个巨型GPU的NVLink开关。
131	To the Vera Bluefield 4 compute that we recently introduced, and I'll say a few words about that later.	Vera Bluefield 4 计算一下我们最近介绍的, 我稍后会说几句话。
132	To Spectrum X, which is our Ethernet solution, making NVIDIA the largest networking company in the world today.	致Spectrum X,即我们的以太网解决方案,使得NVIDIA成为当今世界最大的网络公司.
133	To Grok 3, a decode accelerator for token generation.	给Grok 3,一个解码加速器,用于代号生成.
134	Vera Bluefield 4的存储加速, 我马上会说几句话。
135	And then Vera RTX compute, and I'll mention that in just a second.	然后Vera RTX计算, 我马上会提到这一点。
136	But this now represents the breadth of technology necessary to create the AI factory.	但这代表了创建AI工厂所需的技术的广度.
137	But nothing works unless you have software that activates it all.	但是,除非有软件激活它,否则一切无效。
138	The first part is the system software that runs the AI factory.	第一部分是运行AI工厂的系统软件.
139	These systems are gigantic.	这些系统是巨大的。
140	A 1 gigawatt data center, 1 gigawatt AI factory, costs about $50 billion.	1千兆瓦数据中心,1千兆瓦AI工厂,耗资约500亿美元.
141	Land, power, shell, all of the compute, networking, storage, power, cooling, all connected together, multiple acres of AI factory.	土地,动力,壳体,所有计算,网络,存储,动力,冷却,全部连接在一起,多亩的AI工厂.
142	Just turning it on and operating it is a marvel in itself.	打开和操作它本身就是一个奇迹。
143	The magnificent amount of software necessary to energize this factory, to put it to use and produce tokens is an extraordinary feat.	为了给工厂注入活力, 使其投入使用和生产代币, 所必须的软件量是非凡的功绩。
144	NVIDIA offers the entire full stack software.	NVIDIA提供完整的堆栈软件.
145	On top of that are what we call the CUDAX libraries, the algorithms of our company.	除此之外,我们称之为CUDAX库, 我们公司的算法。
146	What NVIDIA truly is, is an algorithm company that designs silicon to accelerate them.	NVIDIA真正的是, 一个算法公司设计硅加速它们。
147	Of course, one of them is cuDNN, what many researchers recognize as the library that brought forth, that started the big bang of modern AI, cuDNN.	当然,其中一个是CuDNN, 许多研究者都认为它是一个图书馆, 它带来了现代AI,CuDNN的大爆炸。
148	It's cuDNN is to AI what SQL is to the world's database industry.	它的cuDNN是AI什么SQL是世界数据库产业.
149	It's a domain-specific language, a domain-specific library.	它是一个域特有语言,域特有库.
150	Just as OpenGL and DirectX enabled real-time computer graphics, cuDNN enabled the AI computing industry.	正如OpenGL和DirectX启用实时计算机图形一样,cuDNN启用了AI计算行业.
151	350 other libraries, those libraries are in fact the crown jewels of our company.	350个其他图书馆,这些图书馆实际上是我们公司的皇冠珠宝。
152	Everything from computational lithography to signal processing, image processing, data processing, particles, fluids, sparse solvers.	从计算文学到信号处理图像处理数据处理粒子流体零散的解析器
153	Incredible breadth of libraries that many of you in the audience have connected into your company.	图书馆的宽度令人难以置信你们中许多人都与你的公司有联系
154	This is how we open new markets.	这就是我们打开新市场的方式。
155	This is how we connect NVIDIA into algorithms that enable you to do computation at incredible speeds.	这就是我们如何将NVIDIA连接到算法中,使你能够以不可思议的速度进行计算.
156	And then finally on the right-hand side, on that side, the models.	最后在右侧, 在那边,模型。
157	This is the new software, the new engine of new applications.	这是新软件,新应用程序的新引擎.
158	And it sits on top of and it collaborates with the CUDAX libraries.	它坐落在顶端,并与CUDAX图书馆合作.
159	These models NVIDIA have been working on across multiple domains.	这些模型NVIDIA一直在跨越多个领域开展工作。
160	With these models, it opens up new opportunities.	通过这些模型,它开辟了新的机遇.
161	And I have lots to say about that later on.	我以后有很多话要说
162	每一代人 NVIDIA都是从Pascal开始的
163	The AI era started with the Pascal new, what is called DGX1 system.	AI时代始于帕斯卡尔新式,即所谓的DGX1系统.
164	DGX1 brought to the world the world's first AI supercomputer.	DGX1为世界带来了世界上第一台AI超级计算机.
165	I introduced it 10 years ago, delivered the first one to OpenAI, started their computation journey.	我10年前介绍了它, 给OpenAI送了第一个,开始他们的计算旅程。
166	From Pascal to Volta to Ampere, then to many of you who heard about us for the very first time.	从帕斯卡到沃尔塔到安佩尔,然后向你们中许多第一次听到我们的消息的人.
167	Hopper, the current generation Blackwell, and then now Vera Rubin in full production.	霍伯(Hopper),即现代布莱克韦尔,后来薇拉·鲁宾(Vera Rubin)开始全面生产.
168	Notice six generations of architectures.	注意到六代建筑.
169	With each generation, we advanced the technology.	每一代人,我们都会进步技术。
170	In the course of that last 10 years, we've improved AI performance by one million times.	在过去十年中,我们把AI的性能提高了一百万倍.
171	If you would take a step back and think about what would Moore's law have done,	如果你退一步想想摩尔的法律会做什么
172	in 10 years' time, the first five years would have improved by a factor of 10,	10年后,头5年会改善10倍
173	the second five years by another factor of 10, 100x improvement in 10 years' time.	后五年又增加了10倍 10年后100倍
174	In the same period of time, because of the way NVIDIA Extreme Codesigns	在同一期间,由于NVIDIA极端代号的方式
175	were innovating across every single computer, across every single stack of the software,	每一个电脑,每一个软件堆,
176	optimizing new algorithms, creating new algorithms, while creating architectures to accelerate it,	优化新的算法,创建新的算法,同时创建架构来加速它,
177	working across the entire domain, what we call Extreme Codesign,	在整个领域工作, 我们称之为极端代号,
178	we were able to accelerate AI by a million times, and we're going to continue to do so.	我们把AI加速了100万倍,我们将继续这样做。
179	This is what NVIDIA builds today, and this is the foundation by which we can enable Europe	这就是NVIDIA今天所建立的基础,这是我们能够使欧洲得以发展的基础。
180	to be part of the AI ecosystem, the AI economy.	成为AI生态系统,AI经济的一部分.
181	Now, each generation, in the beginning, Pascal was GPUs, Volta was GPUs, Ampere was GPUs,	现在,每一代人,在一开始,帕斯卡尔是GPU,沃尔塔是GPU,安佩尔是GPU,
182	and Hopper was GPUs.	霍普是GPU的员工
183	We invented the new DGX form factor.	我们发明了新的DGX形式因子.
184	Then it was Blackwell that introduced to the world the first rack-scale computing,	之后是布莱克威尔向世界介绍第一个架式计算,
185	and I'll show you that in just a moment.	我马上给你看
186	Rack-scale computing turned one entire system into one giant computer, one GPU,	拉克规模的计算把整个系统变成了一个巨大的计算机,一个GPU,
187	and now, with Vera Rubin, we're extending it to processing beyond models,	而现在,与Vera Rubin, 我们正在扩展它处理超出模型,
188	beyond large language models, to now processing the entire agent.	除了大型语言模型, 现在处理整个代理。
189	We'll refer back to this slide later, but remember this.	我们稍后再参考这张幻灯片,但记住这一点。
190	The agent has to think, the agent has to think, and that's where the model comes in.	代理商必须思考,代理商必须思考,这就是模型出现的地方.
191	Perception, reasoning, planning, and using tools.	观念,推理,规划,以及使用工具.
192	All of that runs on the large language model, and the models are designed to do that.	所有这些都运行在大型语言模型上,模型都是为此设计的.
193	But the agent also has what is called a harness.	但特工也拥有所谓的绳索.
194	The harness is basically the exoskeleton of the digital agent.	牵引装置基本上是数码代理机的外骨架.
195	Just like we have the brain, the rest of our body represents our exoskeleton.	就像我们有大脑一样, 我们身体的其余部分代表我们的外骨骼。
196	The agent is the exoskeleton that orchestrates the work,	经纪人是负责组织工作的外骨骼
197	that allows the agent to actually engage memory, active memory, working memory,	允许代理人实际使用内存,活动内存,工作内存,
198	as well as long-term memory, and activating tools.	以及长期记忆和激活工具
199	Then the third part is a computer that processes knowledge,	第三部分是处理知识的计算机
200	trained for data processing.	接受数据处理培训。
201	And then lastly, a computer dedicated to CUDA-X libraries,	最后是一台CUDA-X图书馆的电脑
202	and its job is to run tools.	它的工作是运行工具。
203	And these computers now represent the entire agentic system.	这些计算机现在代表了整个代理系统。
204	So Vera Rubin is in full production, and it's nice to kind of see it.	所以薇拉·鲁宾已经完全投入生产了,也很高兴能看到它。
205	This is the first computer in the world that we've created.	这是我们创造的世界上第一台电脑。
206	No cables, no wires, completely 100% liquid-cooled, and it's just really delightful.	没有电缆,没有电线, 完全100%的液体冷却, 它只是真的很令人愉快。
207	Hang on.	坚持住。
208	Really delightful to be able to pull it out.	能把它拔出来真令人高兴
209	这是一台Vera Rubin电脑。
210	Okay, so this has two GPUs.	好吧,这有两个GPU。
211	Yep.	对
212	What used to take, in the case of Grace Blackwell, two hours to assemble one node.	在Grace Blackwell案中过去需要两个小时来组装一个节点
213	And when you see it, you'll know what I mean.	当你看到它,你会明白我的意思。
214	Cables and wires everywhere, 80% liquid-cooled.	到处都是电缆和电线,80%的液冷.
215	Now Vera Rubin is 100% liquid-cooled, and it takes less than five minutes to assemble it.	现在Vera Rubin是100%的液体冷却剂,组装它需要不到5分钟.
216	And so, incredible, incredible achievement.	所以,令人难以置信,令人难以置信的成就。
217	Of course, this translates to higher cycle times in manufacturing,	当然,这意味着更高的周期制造业,
218	much, much better reliability, and of course, a lot better performance.	可靠性更高,当然也更好
219	This is the NVLink switch tray, and what this does is it connects 72 of these GPUs.	这是NVLink的开关托盘,它能连接72个这些GPU.
220	There are four of them in here right now.	现在这里有四个人
221	It connects 72 GPUs into one gigantic GPU.	它将72GPU连接为一个巨大的GPU.
222	Okay, so each GPU is 20 petaflops, connecting 72 of them into one gigantic system.	好吧,所以每个GPU是20 petaflops, 连接其中72个到一个巨大的系统。
223	Okay.	摆
224	Thank you.	谢谢
225	And this is our new Vera computer.	这是我们的新Vera电脑。
226	It's okay.	没关系。
227	Go easy.	别紧张
228	Let her, give her a chance.	让她,给她一个机会。
229	Everybody has a chance to say something.	每个人都有机会说点什么
230	But today, let me, no, no, take, just, I'll talk to you later.	但今天,让我,不,不,拿,只是,我会跟你谈谈。
231	I'll talk to you guys later.	我待会再跟你们谈
232	Thank you.	谢谢
233	All important issues.	所有重要问题。
234	But let me get through my keynote.	但让我通过我的主旨发言。
235	We have, we have, so this, this auditorium is packed.	我们有,我们有,所以这个, 这个礼堂是收拾好的。
236	We have a hundred other watch parties around the world.	我们在世界各地还有一百个监视聚会。
237	A hundred watch parties.	一百个看台派对
238	15,000 people are watching us right now.	一万五千人在监视我们
239	And so, I'll get through the keynote.	因此,我会通过主旨。
240	There's some really, really fun things I want to show you in the end.	最后我想给你们看一些非常有趣的东西。
241	Okay, so you just have to, you have to, I have to torture you through some education first.	好吧,所以你只需要,你必须, 我必须折磨你通过一些教育先。
242	And then the entertainment is coming.	然后娱乐即将到来.
243	This, this our brand new, this is our brand new Vera computer.	这是全新的 Vera电脑
244	Vera CPU是一种非凡的CPU兽.
245	Eight instructions per clock.	每时八指示.
246	It has the world's first LPDDR memory system so that we can have high performance, very high bandwidth,	它拥有世界上第一个LPDDR内存系统,这样我们就可以有高性能,非常高的带宽,
247	as well as extremely excellent energy efficient, efficiency.	以及极其出色的节能和效率。
248	The on chip fabric is the fastest of any chip in the world.	芯片织物是世界上最快的芯片
249	As a result, data processing is off the charts.	因此,数据处理不在图表之列。
250	Agentic sessions starting and processing is off the charts.	启动和处理的代理会话不在图中。
251	And this is designed for doing exactly that.	这是完全为了做到这一点。
252	This is our Vera computer.	这是我们的Vera电脑。
253	And this computer here, this computer here, is the Spectrum X.	这台电脑,这台电脑,就是光谱X。
254	This is the world's first co-packaged optics.	这是世界上首个共同包装的光学仪器.
255	And you can see it here, I'll show you.	你可以看到它在这里,我给你看。
256	Some of this, some of this area falls into the ground.	其中有的,有的落在地上.
257	And some of this area you can walk on.	有些地方你可以继续走
258	I'm just not sure which.	我只是不确定是哪个
259	Alright, so, so if I accidentally disappear, you guys know, you guys know us.	好吧,所以,所以,如果我不小心消失, 你们知道,你们认识我们。
260	It's not a stunt.	这不是一个特技。
261	It's an accident.	这是意外
262	And so you can see here, these are the lasers.	你可以看到,这些是激光。
263	And these lasers come up through optics and it directly connects, directly connects to our chip.	这些激光通过光学技术直接连接到我们的芯片
264	And it connects using a process technology we invented with TSMC called COOP.	它使用我们发明的工艺技术和叫做COOP的TSMC连接起来。
265	And so silicon photonics are directly attached right onto our chip.	硅光子直接附着在我们的芯片上
266	Otherwise, what you would see is all of this would be PC board.	否则,你会看到这一切将是PC板。
267	And PC board's transmission capability is much worse than optical cables.	而PC板的传输能力比光学电缆要差得多.
268	And so we now have extremely high data rate, extremely high quality data that comes directly onto the chip.	因此,我们现在有非常高的数据率,非常高的质量数据直接出现在芯片上。
269	And this goes directly into, otherwise transceivers, without going through a PCB and a transceiver goes right into the optics.	这直接进入,否则就是收发机, 不经过PCB,收发机直接进入光学。
270	Okay, so this is the world's first CPO.	好吧,所以这是世界上第一个CPO。
271	We're in complete volume production.	我们正在生产完整的量。
272	And 512 port at 200 gigabits per second.	512个港口每秒200千兆比特
273	The world's most advanced Ethernet switch.	世界最先进的以太网开关.
274	And we're really proud of it.	我们对此非常自豪。
275	Thank you.	谢谢
276	And so, so what that, what that gives us, what that gives us is the ability to process the thoughts, the large language model.	所以,这给了我们什么, 给了我们什么, 给我们处理思想的能力, 大语言模型。
277	Process the agent harness.	处理代理控制器。
278	Run data processing for its knowledge.	运行数据处理以获取知识。
279	And then run accelerated computing for all of the tools.	然后对所有工具进行加速计算。
280	That represents essentially the agent workload.	这主要是代理工作量。
281	And I'll show you more in a second.	我一会儿再告诉你更多
282	While Vera Rubin is in full production and the performance is off the charts.	Vera Rubin正在全面制作中,而表演则不尽人意。
283	Every single generation, as I mentioned, through extreme co-design, we're able to achieve performances that are well beyond reason.	正如我所提到的,每一代人,通过极端的共同设计,我们都能实现远远超出理性的表演.
284	Well beyond what Moore's law would have dictated.	远远超出了摩尔定律的要求
285	And as you know, Moore's law has really largely ended.	如你所知,摩尔的法则已经基本结束了。
286	And so we need these extreme co-design ideas.	所以我们需要这些极端的共同设计的想法。
287	Working from chips to systems, systems to system software to algorithm to the entire AI factory.	从芯片到系统,系统到系统软件到算法到整个AI工厂的工作.
288	Well, Vera Rubin is another giant step on top of Blackwell.	Vera Rubin是布莱克威尔的又一个巨型台阶
289	Blackwell was about 30 times faster than Hopper in doing agentic, in doing large language model processing.	布莱克韦尔在做代理,做大型语言模型处理方面比霍伯快约30倍.
290	And now Vera Rubin is another 30 times greater than Blackwell in just two years time.	而现在薇拉·鲁宾在短短的两年时间里又比布莱克韦尔大30倍.
291	Let's take a look at how Vera Rubin turns into what is a DSX full stack AI factory.	让我们来看看Vera Rubin是如何变成 DSX全堆AI工厂的。
292	AI is creating the largest infrastructure build out in history.	大赦国际正在创建历史上规模最大的基础设施。
293	The factories are massive.	工厂是巨大的。
294	The investments are significant.	投资数额很大。
295	We have to build it right.	我们必须正确地建造它。
296	But great chips alone don't make a great AI factory.	但仅靠伟大的芯片并不能制造出一个伟大的AI工厂.
297	NVIDIA DSX is a full stack AI factory reference architecture, co-innovated with the world's leading infrastructure companies.	NVIDIA DSX是一个完整的堆栈AI工厂参考建筑,与世界领先的基础设施公司共同创新.
298	Compute, networking, storage, software, power and cooling, co-designed as one system.	计算,联网,存储,软件,动力和冷却,共同设计成一个系统.
299	Optimized from design to operation, DSX starts in a digital twin.	从设计到操作的优化,DSX从数字双胞胎开始.
300	Performance, power, water and scale validated virtually before construction begins.	性能、动力、水和规模几乎在施工开始前就得到验证。
301	PTC turns equipment designs into simulation-ready assets.	PTC将设备设计转化为模拟准备资产.
302	DASO Systems connects rack and facility design through model-based systems engineering.	DASO Systems通过基于模型的系统工程连接机架和设施设计.
303	Schneider Electric模拟电力分配和液体冷却.
304	Jacobs brings it all together in the omniverse digital twin.	雅各斯将它汇集在全数数码双胞胎中.
305	Siemens evaluates thermal, acoustic and grid impacts.	西门子对热、声学和电网影响进行评价。
306	And Procore supports construction phase planning.	而Procore则支持建设阶段规划.
307	Power limits how much an AI factory can produce.	动力限制AI工厂能生产多少.
308	DSX Max LPS is an end-to-end control loop that dynamically steers power to match workload demand.	DSX Max LPS是一个端对端的控制循环,动态引导电源以匹配工作量需求.
309	It uses 30% less power for the same output.	它对相同的输出使用30%的功率.
310	DSX rethinks cooling.	DSX重新思考冷却.
311	NVIDIA Vera Rubin systems are designed for closed-loop liquid cooling with coolant temperatures up to 45 degrees Celsius.	NVIDIA Vera Rubin系统是为闭路液冷却而设计的,冷却剂温度最高可达45摄氏度.
312	In suitable climates, outdoor dry coolers release heat into the air without evaporating water.	在合适的气候中,室外干燥的冷却器释放热量到空气中而不蒸发水.
313	At 2 gigawatts, DSX can save 54 million liters of water a day.	在2千兆瓦时,DSX每天可以节省5400万升的水.
314	That's the daily water use of 180,000 households.	这就是18万户人家的日常用水.
315	But an AI factory can take several months to build.	但AI工厂需要几个月的时间才能建成.
316	Until it is operational, it produces no tokens.	在投入使用之前,它不会产生任何符号.
317	Today, we are announcing NVIDIA DSX pod, bringing this full-stack platform into a factory-built, integrated system.	今天,我们宣布NVIDIA DSX舱, 将这个完整的平台带入一个工厂建造的集成系统。
318	它从Vera Rubin NVL72开始.
319	连接17公里Spectrum X电缆的7个液冷计算机架组成了一个Vera Rubin DSX舱.
320	Vera Rubin DSX pods are integrated, fully cabled and factory tested, then delivered in transportable containers.	Vera Rubin DSX舱室进行集成,全线缆和工厂测试,然后用可运输容器交付.
321	Dry coolers replace traditional evaporative cooling.	干冷却器取代传统的蒸发式冷却.
322	800 volts DC connects to the racks for efficient power delivery.	800伏特DC连接到机架以高效供电.
323	Once site utilities are ready, estimated on-site deployment goes from several months to days.	场地设施一旦准备就绪,现场部署估计从几个月到几天不等。
324	NVIDIA DSX AI Factory通过设计负责.
325	And now with DSX pod, AI factories can be built at the speed of light.	而现在有了DSX舱,AI工厂可以以光速建造.
326	More intelligence per watt.	每瓦增加情报
327	Less time to first token.	少一点时间到第一标语。
328	What do you think?	你觉得呢?
329	So, in 2016, we introduced the world's first AI supercomputer.	因此,在2016年,我们推出了世界上第一台AI超级计算机.
330	It was in one node, one GPU node.	它在一个节点, 一个GPU节点。
331	三年前,我们引进了Grace Blackwell NVLink72 架式计算机。
332	Both times, we had to reinvent not just the computer, the software, and how it works, but also the supply chain.	两次,我们不仅需要重塑电脑,软件, 以及它的运作方式, 而且还需要重塑供应链。
333	In the case of NVLink72, we took the entire rack, which would have otherwise been assembled on-site, and we assembled it inside the supply chain.	以NVLink72为例,我们拿走了整个架子,否则这些架子本会就地组装,然后在供应链内组装.
334	We essentially manufactured supercomputers at scale.	我们基本上制造的超级计算机规模。
335	It is the reason why we're able to move so fast and be able to provide for literally several hundred data center build-outs around the world at the same time.	这就是为什么我们能够快速移动并且能够同时提供几百个数据中心的建设。
336	Every single computer manufacturer in the world now manufactures NVLink72.	现在世界上每个电脑制造商都制造NVLink72.
337	The scale at which we are now producing computing is unheard of ever.	我们现在生产计算机的规模是前所未有的。
338	Unimaginable.	不可思议.
339	Well, even then, when you bring these systems onto your sites, there are still months and months of assembly.	即便如此,当你把这些系统带到你的网站上, 仍然有几个月和几个月的组装。
340	And you just heard 17,000 kilometers of wiring to connect just spectrum X for the scale-out computing.	你刚刚听到17,000公里的线条用来连接X频谱用于规模的计算
341	Seventeen kilometers.	17公里长.
342	Seventeen thousand kilometers.	17000公里长.
343	It's a crazy number.	这是一个疯狂的数字。
344	And so, 17 kilometers.	因此,17公里。
345	Incredible numbers.	不可思议的数字。
346	Now, suppose we were to do that all inside the supply chain.	现在,假设我们这样做全部在供应链。
347	Run everything up, build everything up, make sure that it runs completely, get it qualified, burnt in, and by the time we ship it, it is fully functional.	运行所有东西,建立所有东西,确保它完全运行,获得合格,燃烧,到我们发货时,它已经完全运转.
348	The amount of cabling and on-site assembly is going to be greatly reduced.	电缆和现场装配的数量将大大减少。
349	There's something much, much more.	还有很多事情
350	In fact, what's going to happen here is that by moving this into the supply chain, the amount of idle computing, the amount of idle computing that's going to be sitting inside at your data centers will reduce dramatically.	事实上,这里会发生的是, 通过将这个移动到供应链中, 闲置的计算数量, 闲置的计算数量, 将会在你的数据中心里大量减少。
351	Remember this.	记住这一点。
352	Each gigawatt, each gigawatt data center is $50 billion.	每千兆瓦,每千兆瓦数据中心为500亿美元.
353	The revenue generation you expect is tens of billion dollars per year.	你所期望的年收入是数百亿美元。
354	And so, we just said it's $40 billion per year of revenue generation that comes out of that infrastructure.	因此,我们刚刚说,每年从基础设施中产生400亿美元的收入。
355	$40 billion, if delayed by just three months, that's $5 billion.	400亿美元,如果推迟了三个月,那就是50亿美元。
356	Excuse me, what is that?	对不起,那是什么?
357	$10 billion worth of revenues.	收入100亿元.
358	$10 billion worth of revenues will make anybody anxious.	价值100亿美元的收入将使任何人感到焦虑。
359	And so, by moving this into the supply chain, the speed by which we can turn on, go from installation, arrival of computing, to production of tokens will go from many months, we think something along the lines of four to five months down to about a month.	因此,通过将它转移到供应链中, 我们可以打开它的速度, 从安装,计算机的到来, 到制作信物, 将会从许多月, 我们想到一些东西, 从4到5个月到大约1个月。
360	Okay, so this is a complete transformation of the world supply chain, and this will change the way that people install and build data centers.	好了,这就是世界供应链的彻底转变, 这将改变人们安装和建设数据中心的方式。
361	And hopefully, as a result, velocity goes up, the wasted idle compute, the wasted idle CAPEX will reduce, and everybody becomes more profitable.	希望,因此,速度会上升, 浪费的闲置计算, 浪费的闲置CAPEX会减少, 每个人都会更赚钱。
362	And so, this is MVLink 72 now into a large cabinet system.	所以,这是MVLink 72 现在进入一个大型的内阁系统。
363	We call it DSX pod.	我们称之为DSX舱。
364	Okay, so this is the next click of NVIDIA computing.	好了,这是NVIDIA的下一击。
365	Now, what you make and how you make it makes you special.	现在,你做的是什么,你是如何让你特别的。
366	Just as here in Europe, you, the work you know makes you special.	就像在欧洲,你,你知道的工作让你特别。
367	You design machines, you operate factories, you discover medicine, you have extraordinary craft.	你设计机器,经营工厂,发现医学,拥有非凡的工艺.
368	That craft, that specialty makes you special.	那个手艺,那个专业让你特别
369	Well, what makes NVIDIA special are really two things.	那么,是什么使得NVIDIA特别是真正的两件事。
370	The first thing is that our architecture is the only accelerated computing architecture that is general purpose.	首先,我们的建筑是唯一一个通用的加速计算建筑。
371	We don't specialize on just one algorithm, for example, transformers.	我们不专门研究一个算法,例如变压器。
372	We specialize across anything that could be paralyzed.	我们专门研究任何可能瘫痪的东西。
373	Could be fluids and particles, it could be quantum, it could be quantum chemistry, could be image processing, could be data processing, structured vector data processing.	可以是流体和粒子,可以是量子,可以是量子化学,可以是图像处理,可以是数据处理,是结构化的矢量数据处理.
374	Processing of all kinds.	各种处理.
375	And because we've been working at it for so long, and because NVIDIA's architecture has remained consistent for literally two and a half decades.	因为我们工作了这么长时间, 并且因为NVIDIA的建筑已经维持了25年。
376	The number of algorithms, the number of domains, the number of industries we serve, we accelerate, has grown tremendously.	算法的数量,域的数量, 我们服务的产业的数量,我们加速, 已经取得了巨大的发展。
377	Our install base, our general purpose architecture, our dedication to this architecture, and our collaboration with all of the world's industries	我们的安装基地,我们的通用建筑,我们对这一建筑的奉献,以及我们与世界所有工业的合作
378	has made NVIDIA's architecture literally everywhere.	已经使NVIDIA的建筑几乎无处不在。
379	The second thing is how we do it.	第二件事是我们怎么做到的。
380	Of course, we have the architecture that we developed, but we also build it now in a full stack, complete AI factory.	当然,我们有我们开发的建筑, 但我们现在也把它建在一个完整的堆栈,完整的AI工厂。
381	There are many companies in the world who want to design their own chips.	世界上有很多公司想自己设计芯片.
382	But most of the companies in the world just want to deploy the AI infrastructure so that they can run their business on top of it.	但是,世界上大多数公司只是想部署AI基础设施,以便他们能够经营自己的业务。
383	The ability to design a chip and build a supercomputer, write all the software, is really completely out of mind for most of the companies around the world.	设计芯片和建造超级计算机的能力,写出所有软件的能力,对于世界上大多数公司来说,其实是完全疯了.
384	And so we now have the combination of these two ideas, CUDA, full stack AI factories, and it exposes us to markets that otherwise isn't possible.	现在我们有了这两个想法的结合, CUDA, 完整的堆积AI工厂, 它让我们暴露在市场中, 否则是不可能的。
385	We're able to partner with so many companies here for this very reason.	为了这个原因,我们能和这么多公司合作
386	If I were to go around this, starting with CSPs, this is where most people think the vast majority of AI is today.	如果我绕过这个,从CSP开始,这是大多数人认为今天绝大多数AI的地方.
387	But as you can see in the beginning of our keynote, I mentioned a whole bunch of companies who are not those.	但正如你们在主旨发言开头所看到的, 我提到了一大堆公司,它们不是那些公司。
388	So CSPs remain very large businesses for the world's computing.	因此CSP仍然是世界计算学的非常大的企业.
389	But there are AI natives, companies that are starting all over the world, there are so many that are here,	但有AI本地人, 公司开始在世界各地, 有这么多在这里,
390	Neoclouds, because every region wants to participate in the AI economy.	Neoclouds,因为每个地区都希望参与AI经济.
391	Every region wants to ensure that the valuable resources that they have, the energy that they have, is dedicated to companies in their region.	每一个区域都希望确保它们所拥有的宝贵资源,以及它们所拥有的精力,都献给本地区的公司.
392	For example, Mistral here in Europe, Mistral in France, Nebius here in Europe, so many companies around Europe that are building up Neoclouds.	例如,欧洲的米斯特拉尔,法国的米斯特拉尔,欧洲的尼比乌斯, 如此多的欧洲公司正在建立新星。
393	OEMs and ODMs who are building computers for the world's enterprise.	为世界企业建造计算机的OEMs和ODMs.
394	AI labs, of course we know about last year, all we spoke about was OpenAI.	AI实验室,我们当然知道去年,我们谈论的只是OpenAI。
395	今天我们谈论OpenAI,Anthropic, 但我们也谈论不可思议的Metasmuse, Grokbot, 当然还有双子座。
396	So many different AI labs are now participating at the frontier.	现在有很多不同的AI实验室在边境参与.
397	And then very importantly, and this is ultimately the most important,	这才是最重要的
398	for AI to succeed, for Europe to benefit from AI and to have really created an AI ecosystem and AI economy,	让AI成功,让欧洲从AI中获益,真正创造了AI生态系统和AI经济,
399	you need all of the end markets, the end industries to adopt AI and to succeed with AI.	您需要所有终端市场,终端行业采用AI,并成功AI。
400	Whether it's in the automotive industry, because we're building self-driving cars, or the way you manufacture with robotics,	无论是在汽车工业, 因为我们正在建造自驾汽车, 还是你制造机器人的方式,
401	the way you discover new drugs in healthcare, your financial services industry,	你发现新药物的方式在医疗保健,你的金融服务行业,
402	all of these different industries have to become AI companies, every single one of them.	所有这些不同的产业都必须成为AI公司,每一个都成为AI公司.
403	Well, we have a unique ability through what I just mentioned about CUDA and the way that we build computers.	通过我刚才提到的CUDA和我们制造计算机的方式,我们拥有独特的能力。
404	We have a unique ability to serve literally every single segment of AI.	我们拥有独特的能力,从字面上为AI的每一个部分服务.
405	And so one architecture, many markets, and because these infrastructures are so significant and so expensive,	因此,一个建筑,许多市场, 由于这些基础设施是如此重要如此昂贵,
406	we need to make sure that they're as productive as possible, as soon as possible,	我们需要确保它们尽可能有生产力尽快
407	that they're fungible, that you can use it for every workload, from the beginning of your AI journey,	他们可以互换, 你可以使用它每一个工作量, 从开始你的AI旅程,
408	data processing to training, to post-training, inference and agentic processing, every single step of the AI journey,	数据处理到培训后培训推论和代理处理
409	to every type of AI, chemicals and physics, to languages, to every single industry and the algorithms that they use.	各种类型的AI,化学和物理,语言, 每一个行业和它们使用的算法。
410	And so NVIDIA's architecture is fungible, and as a result of that, continuous innovation, continuous updating of software,	因此NVIDIA的架构是可互换的,因此,不断创新,不断更新软件,
411	because the install base is so great, we have such an incredibly rich ecosystem of software,	因为安装基础如此之大我们拥有如此丰富的软件生态系统
412	then these architectures are also durable, fungible, durable, productive.	那么这些建筑也是耐久的,可替换的,耐久的,有生产力的.
413	Well, that lays the foundation.	这奠定了基础。
414	That is the infrastructure.	这就是基础设施。
415	Now what do we do with it?	现在怎么办?
416	This last year, we saw a transition in AI.	去年,我们看到了AI的转变.
417	AI is now productive.	AI现在很有成果。
418	And one of the most important times was in the beginning of the year when we spoke about OpenClaw,	其中最重要的一次是在年初我们谈到OpenClaw时
419	which introduced the concept of an open source agent, and it really started what we all see across the world today.	它引入了开源代理的概念, 它真正地开始了我们今天所看到的。
420	Agents think, reason, iterate, use tools, and as a result, consume an enormous amount of tokens.	特工们思考,理性,仪表,使用工具,结果消耗了大量的代币.
421	But as a result, they become useful.	但因此,它们变得有用.
422	So more agents are being developed, they're more useful,	所以,更多的代理正在开发, 他们更有用,
423	and therefore they're used by more people, and because they have to use a lot of tokens to think,	因此他们被更多的人使用因为他们必须用很多符号来思考
424	the amount of token consumption has gone through the roof.	象征性消费的数量已经从屋顶上走过。
425	GPU demand has really gone through the roof.	GPU的需求真的突破了屋顶.
426	Now what's processing it is this.	现在处理的是这个。
427	This is an agent.	这是特工
428	These are the five elements of an agent.	这些都是五要素一个代理。
429	The first part are the language models.	第一部分是语言模型。
430	These are the tokens that are running on top of the GPUs,	这些是运行在GPU上面的标志,
431	and they're generating tokens that are consumed by the agent, what I call the exoskeleton.	他们正在生成代币被毒剂消耗, 我称之为外骨骼。
432	That OpenClaw harness sits inside a sandbox, and that sandbox isolates it from the rest of the system.	OpenClaw的绳子坐落在沙盒里沙盒将它与系统的其他部分隔离开来
433	This is a very important concept that we're going to come back to.	这是一个非常重要的概念,我们将回到。
434	There are five things.	有五义.
435	There's the model, there's the harness, the sandbox that the harness sits in,	有模型,有绳子, 沙盒,绳子坐在那里,
436	and two other things, knowledge and tools.	还有另外两件事知识和工具
437	Models, sandbox, harness, knowledge and tools.	模型、沙盒、绳索、知识和工具。
438	These are the five elements of an agent.	这些都是五要素一个代理。
439	Now before I talk about how to make the agents useful,	在我讲到如何让特工有用之前
440	the most important thing is we have to make sure that the agent is secure.	最重要的是我们必须确保特工安全
441	The agent is autonomous in this autonomous system that has access to files, has access to resources, networks,	代理在这种自主系统中是自主的,可以访问文件,可以访问资源,网络,
442	access to tools, and we need to make sure that the privileges that we provide this autonomous system is minimal	获取工具,我们需要确保我们提供的这种自主系统的特权是最低限度的
443	and that we constantly monitor it.	并且我们一直监视着它
444	We introduced a fantastic new product, a new platform, and it's called OpenShell.	我们推出了一个奇妙的新产品,一个新的平台,它叫做OpenShell.
445	Openshell来得正好
446	This is a secure runtime that provides for the things that I just mentioned	这是一个安全的运行时间提供的东西我刚才提到
447	and has an additional hardware-based monitor.	并拥有额外的基于硬件的显示器.
448	Before I explain it to you, let's watch a video.	在我给你解释之前,让我们看看视频。
449	Prizes are preparing to deploy billions of AI agents.	奖项准备部署数十亿AI特工.
450	To be useful, they'll reason, plan, use tools, access data, and work together.	为了有用,它们会理性,计划,使用工具,访问数据,并合作.
451	Today, the only thing stopping an agent from misusing all that is asking it nicely.	今天,唯一能阻止特工滥用一切的就是善意的要求
452	Please don't delete anything.	请勿删除任何东西。
453	Please don't push to main.	请不要推到主。
454	Please don't touch my SSH keys.	请不要碰我的SSH钥匙。
455	That's the please don't approach.	请不要靠近
456	But hope isn't security.	但希望不是安全的。
457	Safety isn't an aspiration.	安全不是一个愿望。
458	It's technology and engineering.	科技与工程.
459	NVIDIA Open Shell和哨兵将这一哲学付诸实践。
460	You write a policy for what each agent can touch, and OpenShell enforces it at the granularity of system calls, file access, and network connections.	您为每个代理能够触摸的东西写一个策略, OpenShell 在系统调用、文件访问和网络连接的颗粒性上执行它。
461	Formal verification checks every policy change against your permission ceiling, so agents never gain more access than intended.	正式核查根据你的许可上限检查所有政策变动所以特工们从来就得不到比预期更多的访问机会
462	It's like the early internet.	这就像早期的互联网。
463	Viruses were everywhere until browsers enforced strict limits on what a web page could do.	病毒无处不在,直到浏览器对网页所能做的事情实施严格的限制.
464	OpenShell对特工也是这样
465	Enforcement happens outside the agent, not just in LLM guardrails, bringing agent-native security into your infrastructure.	执法发生在特工之外不只是在LLM守卫设施里给您的基础设施带来代理安全
466	Every agent runs inside its own isolated sandbox, with no direct route to the network.	每个特工都运行在自己孤立的沙盒内,没有直接通往网络的路线.
467	Every outbound network request goes through the supervisor, outside the sandbox.	每一个外出网络请求都通过主管,在沙盒之外.
468	It checks the action against your policy.	它检查行动与你的政策。
469	If it's not permitted, it's denied.	不许者,即绝也.
470	Keys stay outside, with the supervisor, and are added only on the way to an approved endpoint.	钥匙留在外面,与主管一起,只在经过批准的终点处的路上才增加。
471	A compromised agent can't leak a key it never had.	失密的特工不能泄露它从来没有过的钥匙
472	The gateway is the control plane.	网关是控制飞机.
473	The supervisor makes the decisions.	由主管决定。
474	The sandbox holds the agent, and the policy prover checks proposed policies against your permission ceilings.	沙盒持有代理, 政策证明人检查拟议的政策与您的许可上限。
475	OpenShell combines Linux kernel isolation with runtime policy enforcement.	OpenShell将Linux内核隔离与运行时政策执行相结合.
476	NVIDIA Sentry adds an optional layer of protection outside the host itself, away from the agent.	纳米比亚哨兵在宿主本身之外,远离代理人,增加了一个可选的保护层.
477	Running on NVIDIA BlueField 4, Sentry monitors agent activity and enforces security policies in a hardware-isolated trust domain.	在NVIDIA Blue Field 4上运行,哨兵监视代理活动,并在硬件同化的信托域执行安全政策.
478	Because agent reasoning traffic passes through Sentry, if the agent is planning to do something unintended, it will update policies immediately and cut the connection to the model, quarantining it in milliseconds.	因为代理推理流量通过哨兵,如果代理正在计划做一些意料之外的事情,它会立即更新政策,并切断与模型的连接,以毫秒的速度断绝.
479	And it keeps tamper-evident records that operators can verify for detection, investigation, and response.	它不断篡改不言自明的记录操作人员可以核实用于侦测、调查和反应
480	OpenShell is open source under Apache 2.0, runs everywhere your agents do, edge to cloud, and takes two commands to adopt.	OpenShell在Apache 2.0下是开放源代码,运行到您的代理处,边缘到云,并需要两个命令才能通过.
481	Install the CLI, create a sandbox, and run your favorite agent.	安装 CLI, 创建一个沙盒, 运行您最喜欢的代理。
482	Agents are only as useful as the access you give them.	特工们和你给他们的准入一样有用
483	OpenShell and Sentry put that access under your control.	OpenShell和哨兵把进入权交给你
484	That's how we advance AI capabilities and safety together.	这就是我们一起推进AI能力和安全的方法.
485	Agents are autonomous systems.	特工是自动系统
486	We have to create extremely secure technology to make sure that we could develop these agents and deploy these agents safely.	我们必须建立非常安全的技术,以确保我们能够发展这些代理人,安全部署这些代理人。
487	Now, it's incredible.	现在,这是不可思议的。
488	After announcing OpenShell, just about every major company in the world, from the Frontier Labs to the operating system companies, cybersecurity companies, platform companies, have come out to adopt OpenShell.	在宣布OpenShell之后,仅仅世界上每个大公司,从边境实验室到操作系统公司,网络安全公司,平台公司,都出来采用OpenShell.
489	It's really fantastic to see.	这真是太棒了。
490	And the reason for that is because this technology has arrived at exactly the right time.	之所以如此,是因为这个技术已经到达了正确的时机。
491	At a time when we want to deploy agents all over the world.	此时我们想在世界各地部署特工。
492	Now remember, this isn't just about the services using OpenShell as they deploy agents.	现在记住,这不仅仅是服务使用OpenShell 他们部署代理。
493	This is really about all of the world's companies also adopting OpenShell so that you could receive these agents and have them run autonomously inside your company.	这是关于世界上所有的公司也采用OpenShell 这样你就可以接收这些代理让他们在你的公司里自主经营。
494	Every single level of capability, every single task that you give the agent for every person that uses the agent, every one of those combinations requires a different level of privilege that you would provide to the agent.	每一级的能力,你给代理人的每一项任务每一个使用代理人的人,每一个这些组合都需要你给代理人提供不同的特权
495	And every single time an agent is launched, the privileges are different.	而每次推出代理时,特权都不同.
496	And so privileges, the policy enforcement, the monitoring, all has to be done dynamically.	因此,特权,政策执行,监控, 都必须动态地进行。
497	所以OpenShell是真正的革命。
498	I'm delighted to say that Anthropic and NVIDIA are partnering to create cloud OpenShell agents.	我很高兴地说,Anthropic和NVIDIA正在合作创建云开壳代理.
499	And so you can go to the website.	然后,你可以去网站。
500	There's a really great write-up about that.	写得很好
501	所以,第一部分,第一部分是如何使用这些带子, OpenClaw, Hermes, OpenHand, 你可以从Langchain, DeepAgents, 或者用Python来写。
502	There's a whole bunch of different ways you could create agent harnesses that sits inside now an OpenShell secure harness.	有一大堆不同的方式你可以创造特工的带子现在坐在里面一个OpenShell安全带。
503	Okay, and that's the sandbox that it sits in, the secure sandbox.	好吧,这就是它坐在那里的沙盒,安全沙盒。
504	And then next, let's talk about models.	然后,我们来谈谈模型。
505	You should use models of all kinds.	你应该用各种模型
506	In the future, it won't just be one model that we run everything on.	在未来,它不会只是一个模型我们运行的一切。
507	We're going to run, we're going to use the best models whether they're closed, open, or custom.	我们要运行,使用最好的模型无论它们关闭,开放,还是习惯。
508	We're going to use all the models to do your work.	我们要用所有的模型来完成你的工作。
509	Sometimes the work that you do is fairly generic.	有时候你的工作很普通
510	Artificial general intelligence is going to do an incredibly good job.	人工一般情报工作会做得非常好
511	However, sometimes the work that you want to do captures too much of your proprietary knowledge, too much of your specialty, and you want to continuously enhance that intelligence.	然而,有时你想要完成的工作会捕捉到太多的专有知识,太多的专长,你想要不断增强这种智力.
512	Every company in the future will use both closed models as well as open models that you customize.	未来每个公司都会同时使用封闭模型以及您定制的开放模型.
513	By the end of the session, you will see that, in fact, building your own AIs is not that hard.	到会议结束时,你会看到,事实上,建立自己的AI并不是那么难.
514	It starts with, of course, the model.	当然是从模型开始的
515	We've been dedicated to advancing open models to open up new opportunities and enable people to build their own custom AIs.	我们致力于推进开放模式,开辟新机遇,使人们能够建立自己的定制AI.
516	Nemotron is for languages.	尼莫特龙是语言
517	Cosmos is our world foundation model.	宇宙是我们的世界基础模型。
518	We are the frontier.	我们是边界。
519	Groot for human robotics.	人类机器人的格鲁特
520	We are the frontier.	我们是边界。
521	Ising for autonomous navigation.	进行自主导航。
522	We are the frontier.	我们是边界。
523	Ising for quantum classical computing.	Ising用于量子古典计算.
524	We are the frontier.	我们是边界。
525	And, of course, Earth-2, physics, AI.	当然还有地球二号物理 AI
526	We are the frontier.	我们是边界。
527	So, in all of these different areas, we offer you all of the models, and we're continuously working on it.	因此,在所有这些不同的领域, 我们向你们提供所有的模型, 我们正在不断研究。
528	We've been working on it for a decade, and we'll continue to work on this forever.	我们为此努力了十年之久,我们将永远继续为此而努力.
529	We offer you the model completely open, the data that goes along with it so that you know the providence of the model that we created,	我们为你们提供了完全开放的模型, 随附的数据,让你知道我们创造的模型的本能,
530	as well as the scripts necessary for you to reproduce it.	还有你复制它所必需的剧本
531	So, not only do you have the open weights, you have from us the entire recipe and the data to recreate the open models	所以,你不仅有开放的权重, 你从我们那里得到了整个食谱和数据来重现开放的模型
532	so that you could know where it came from and have complete trust into it.	以便你们知道从哪里来的,并完全信任它。
533	Well, these models are not necessarily the frontier of frontiers, especially for the language model.	这些模型不一定是前沿的特别是语言模型
534	However, with customization and with partners that the ecosystem that we work with, you could turn these models into frontier.	然而,随着我们的生态系统的定制化和合作伙伴,你可以把这些模型变成前沿。
535	And, as a result, turn your proprietary data into sovereign intelligence.	结果,把你的专有数据变成主权情报
536	Here's an example of A21 helping Alice turn Nemotron into a model that exceeds frontier capabilities.	以下是A21帮助爱丽丝将尼莫特龙变成超越前沿能力的模型的例子.
537	Here's Base 10 helping CodeRabbit turning a Nemotron model, after fine-tuning on your data, turned into frontier capability, Base 10 and CodeRabbit.	这是10号基地帮助CodeRabbit转换一个Nemotron模型在微调你的数据后变成了前沿能力 10号和CodeRabbit
538	Here's Trajectory and Harvey, again using Nemotron, post-trained it and turned and created their own frontier-level capability.	这是Trajectory和Harvey,再次使用Nemotron,后训练它,并转身并创建了自己的前沿能力.
539	这是Skydance的Scale公司再次做同样的事情
540	It goes on and on and on.	随顺不断.
541	The point being that you could start with an open model, something that's incredibly good, like Nemotron,	重点是你可以开始一个开放的模型, 一些令人难以置信的好, 像尼莫特隆,
542	and with your own data, post-train it for your use case, and achieve, of course, super intelligence.	并且用你自己的数据, 后训练它为您的使用案例, 当然,实现超级智能。
543	That is a very sensible thing to do, and all the technology exists today.	这是一件非常明智的事情,所有的技术今天都存在。
544	Here's an example of Palantir with Tampa General Hospital.	以下是帕兰蒂尔与坦帕总医院的一个例子.
545	The list goes on and on.	名单不断.
546	We have a rich ecosystem of partners that can help you, and they're incredible companies, and they could work with you from around the world.	我们有一个丰富的生态系统合作伙伴可以帮助你, 他们是不可思议的公司, 他们可以与你合作从世界各地。
547	I'll mention one and I'll offend everybody else.	我会提到一个,我会冒犯其他人。
548	I'll have to sit here and mention all of them.	我不得不坐在这里提到他们的全部。
549	I love them all.	我爱他们
550	First, agent harness running in a secure sandbox open shell.	首先,特工的吊带运行在一个安全的沙盒打开壳。
551	Second, Nemotron large language models, the first two components of what you need.	第二,尼莫特隆大语言模型,你需要的前两个组件.
552	But ultimately, what turns these agents into super agents is that you give them access to your company's knowledge.	但归根结底,这些特工变成了超级特工就是你让他们了解你的公司
553	And this is really special.	这真的很特别。
554	This is called the Agent Knowledge Foundation that we put together.	这叫"特工知识基金会" 我们一起建立的
555	Libraries and models that are completely world-class.	图书馆和模型都是世界级的
556	It allows you to turn your company's data, the data could be in, the assets could be in unstructured format in documents,	它允许你调用你公司的数据, 数据可能是在, 资产可以是非结构格式的文件,
557	so that you need an AI to understand how to read these PDFs and read all these unstructured documents, or it could be in structured data inside databases.	因此,你需要一个AI来理解如何读取这些PDF并读取所有这些无结构的文档,或者可以在数据库内部的结构化数据中读取.
558	We have the libraries and the tools necessary to help you extract meaning and understanding from your company's data.	我们拥有必要的图书馆和工具来帮助你从公司的数据中获取意义和理解.
559	One at a time, Nemotron parse, completely world-class top of leaderboard.	一次一个,尼莫特龙完全世界一流的领导板。
560	Nemotron embed, taking a query and associating with the knowledge that's inside the vector database.	Nemotron嵌入, 接受一个查询并与在矢量数据库中的知识联系起来。
561	Top of leaderboard.	上头板.
562	Nemotron re-rank, it might find many different sources where the query is associated with different pieces of data.	Nemotron重新排序,它可能会发现许多不同的来源,其中查询与不同的数据块相关.
563	It brings it together, ranks them based on which one's the most relevant to you.	它将它聚集在一起, 排序他们的基础一个最适合你。
564	Kumo, to be able to predict from your structured database what is likely to happen.	库莫,从你结构化的数据库里能够预测会发生什么
565	And so this is, for example, if you were to receive a notice that one of your suppliers will be short on supply or delay on supply,	因此,举例来说,如果你收到一个通知, 你的供应商将短缺供应或延迟供应,
566	that piece of information is likely unstructured data as a document.	作为文档,该信息可能是非结构数据。
567	The information about the orders in your company are likely inside a structured database, your SQL database, for example.	有关您公司订单的信息可能存在于一个结构化的数据库,例如您的SQL数据库.
568	And in order to predict what's going to happen to all of these orders who are going to be short, the amount of delay, you use Kumo to make that prediction.	为了预测这些订单会怎样, 这些订单将会是短暂的, 延迟的数量, 你用Kumo来预测。
569	And then in order to figure out what to replace it with, you might use vector database search and go find an engineering document of a replacement component.	然后,为了找出用什么来替换它, 你可以使用矢量数据库搜索, 并去找一个工程文件一个替换组件。
570	Okay, so all of those different libraries and data processing capabilities now makes for you your knowledge foundation.	好了,所以所有这些不同的库和数据处理能力现在让你的知识基础。
571	You've now armed your AI with proprietary data, proprietary knowledge.	你现在已经武装你的AI 拥有专有数据,专有知识。
572	Now, the last part they'll talk about is ultimately tools.	现在,他们要谈的最后一部分是工具。
573	Tools of the trade.	交易工具。
574	NVIDIA has been working on CUDA-X, and for most of you, before AI, all of our conversations have been around how do we accelerate simulators and solvers,	NVIDIA一直在研究CUDA-X,对你们大多数人来说,在AI之前,我们所有的对话都围绕着我们如何加速模拟器和解析器,
575	data processing engines, domain-specific libraries of all kinds.	数据处理引擎,各类域特定库.
576	They're used in scientific discovery in all the different fields of science.	它们被用于所有不同科学领域的科学发现.
577	They're used in engineering.	它们被用于工程.
578	They're used in research.	它们被用于研究.
579	They're used in designing, building creative arts.	它们被用于设计,建设创作艺术.
580	At this core are CUDA-X libraries, and these CUDA-X libraries are algorithms.	在这个核心是CUDA-X库,这些CUDA-X库是算法.
581	And algorithms, as you know, are beautiful.	而算法,你知道,是美丽的。
582	Let's take a look at them.	让我们看看他们。
583	Math is beautiful.	数学很美丽
584	CUDA-X libraries, these are the crown jewels of our company, and the reason for that is this.	CUDA -X图书馆这是我们公司的皇冠珠宝原因就是这个
585	When you're building an AI assistant, it could be a lab assistant, it could be a research assistant, it could be a co-designer,	当你建造一个AI助理, 它可以是实验室助理, 它可以是研究助理, 它可以是联合设计师,
586	as somebody who's co-creator with you, building something with you.	作为你共同创作的人, 与你一起建造一些东西。
587	Ultimately, what makes those agents useful, what makes them able to do real work, are the tools.	归根结底,是什么使这些代理人有用,是什么使他们能够做真正的工作,就是这些工具。
588	And so we now have models that allow these agents to reason, understand information, reason, to be able to plan and iteratively act and use tools until it could get the job done.	因此,我们现在有了模型, 允许这些代理商理性,理解信息,理性, 能够规划和反复行动, 使用工具,直到它能够完成工作。
589	But it still comes down to the tools, and these tools we've been working on for so many years.	但是,它仍然下到工具, 而这些工具我们已经工作了这么多年。
590	And there was one particular computer I showed you on the right-hand side that runs every single one of these tools, and it could be fluid dynamics and particle physics.	我给你们看的右手边有一台电脑运行这些工具中的每一个, 它可以是流体动力学和粒子物理学。
591	It could be quantum chemistry.	可能是量子化学
592	It could be data processing.	这可能是数据处理。
593	It could be ray tracing.	可能是射线追踪
594	All of these tools run on that one computer, and that one computer speeds up the processing of it.	所有这些工具都运行在一台计算机上, 而一台计算机可以加速它的处理。
595	compared to just running it natively on a CPU,	而不是在CPU上用本地语言运行
596	由10,50,100x。
597	And CUDAX tools is what give the agents the tools of the trade.	而CUDAX工具是给予代理商交易工具的.
598	And it's so cool to see how other people are using it.	看别人怎么用它,太酷了
599	Let's take a look.	让我们来看看。
600	So this is the magic, magic behind the agents.	这就是特工背后的魔法
601	Ultimately, the agents have to use tools.	归根结底,特工必须使用工具。
602	Now, of course, in one of the first fields of agentic use	现在,当然,在最早的药剂使用领域之一
603	is software programming.	是软件编程。
604	And therefore, a compiler or a Python is all really it needs.	因此,编译器或Python是它真正需要的。
605	But in the future, as we have these tools do more and more	但在未来,我们有这些工具做越来越多的
606	and more and more useful work, we're	和越来越多的有用的工作,我们
607	going to need to give them more tools.	需要给他们更多的工具。
608	And here's some examples.	这里有一些例子。
609	So here's Lyta Sciences.	这就是莱塔科学。
610	They're creating a lab assistant.	他们正在建立一个实验室助理。
611	他们使用NVIDIA的BioNemo模型
612	that are designed for understanding proteins	用来理解蛋白质
613	and understanding cells and generating proteins	理解细胞和生成蛋白质
614	and chemicals in drug discovery.	和药物发现中的化学品。
615	他们使用BioNemo和CUDAX工具.
616	This is Siemens Factory, building a factory agent	这是西门子工厂,建造一个工厂代理
617	that uses NVIDIA Cosmos, our world foundation	我们的世界基金会
618	model that understands the physical world and CUDAX tools.	能够理解物理世界和CUDAX工具的模型。
619	And here's an example of Dassault	这是达索的一个例子
620	creating a virtual companion that	创建虚拟伴侣
621	使用NVIDIA的Nemotron, 我们的语言模型,
622	还有CUDAX工具
623	Let's see them in use.	让我们看看他们正在使用。
624	The most intelligent scientist still needs a laboratory.	最聪明的科学家仍然需要一个实验室.
625	The best engineer still needs simulators and solvers.	最好的工程师仍然需要模拟器和解析器.
626	An AI agent is no different.	人工智能特工也一样
627	An agent brings together five building blocks.	一个特工集中了五个构件
628	The model perceives reasons and plans.	模型能感知原因和计划.
629	The harness coordinates the work.	绳子协调工作
630	Knowledge foundation gives the agent context	知识基础提供了代理商的背景
631	from your company's structured and unstructured data.	从你公司的结构化和无结构化的数据。
632	Tools let the agent simulate, experiment, and act.	工具让代理模拟,实验,并采取行动.
633	NVIDIA Open Shell控制着哪些代理可以访问和执行.
634	Start with a frontier closed model	从前沿封闭模型开始
635	or customize an open model.	或自定义打开的模型。
636	With NVIDIA's agent toolkit, any company	NVIDIA的代理工具箱任何公司
637	can turn its knowledge and expertise	能够改变其知识和专长
638	into a customized super agent.	成为定制的超级特工
639	Let's look at three examples.	略举三例.
640	对Lyla Sciences来说目标是设计一种疗法
641	for B-cell lymphoma.	B细胞淋巴瘤
642	Lyla Iris的代理商在NVIDIA尼莫特龙之上
643	turns that goal into an experimental plan.	把目标变成实验计划
644	In the computational dry lab, Lyla agents	在计算干燥实验室,Lyla代理
645	use NVIDIA BioNemo Agent Toolkit	使用 NVIDIA 生物内莫剂工具包
646	to call models for protein design and RNA optimization,	可以调用蛋白质设计和RNA优化模型,
647	running millions of virtual experiments.	运行着数百万个虚拟实验
648	The most promising candidates move	最有前途的候选人流动
649	into Lyla's AI Science Factory,	进入莱拉的AI科学工厂,
650	where autonomous robots and instruments,	其中自主机器人和仪器,
651	由NVIDIA Omniversis和宇宙公司开发,
652	make and test them.	做测试
653	The results return to Iris, shaping the next designs	结果是返回到Iris,塑造下一个设计
654	and experiments.	和实验。
655	The loop continues.	循环继续。
656	At Siemens Electronic Factory Erlangen,	在西门子电子工厂埃兰根
657	the goal is to bring new products into production.	目标是将新产品投入生产。
658	Process preparation ex-agents use NVIDIA Cosmos	流程准备前代理人使用 NVIDIA 宇宙
659	to generate virtual prototypes and synthetic data,	生成虚拟原型和合成数据,
660	then train and validate inspection models	然后培训和验证检查模式
661	before production starts.	在开始生产之前。
662	Intelligence Center ex-agents,	情报中心前特工
663	built with NVIDIA's Factory Operations Blueprint,	与NVIDIA的工厂业务蓝图一起建造,
664	use orders, supplies and line setups	使用订单、用品和线路设置
665	to recommend production plans.	建议生产计划。
666	数字双人编译器使用 NVIDIA Omniverse 图书馆
667	to validate those plans with physics simulation	通过物理模拟验证这些计划
668	and real-time factory data	和实时工厂数据
669	before changes reach the factory floor.	在变化到达工厂地板之前。
670	With the Dassault Systems platform,	与达索系统平台,
671	an engineer needs to adapt a Volkswagen Golf Door design	工程师需要适应大众汽车高尔夫门的设计
672	for the Tiguan.	为提瓜人.
673	Aura 和 Leo 使用 NVIDIA 奈马特龙
674	to reason over engineering requirements.	在工程要求上讲道理
675	The harness coordinates the specialized sub-agents	带子协调专业的副代理
676	and preserves context across planning, design and simulation.	并保存整个规划、设计和模拟的背景。
677	Aura用Inovia来策划工作
678	around the vehicle's dimensions, weight limits	围绕车辆的尺寸、重量限度
679	and other engineering requirements.	和其他工程需求。
680	Leo Design Agent selects a compatible Golf Door design	狮子座设计代理选择兼容的高尔夫门设计
681	and uses Catia to reshape it for the Tiguan.	并用卡提亚重塑它为提瓜人。
682	Leo Simulation Agent uses Simulia,	里欧模拟特工用西穆利亚
683	由NVIDIA CUDA-X图书馆加速,
684	to test the variance, revealing a vibration problem.	以测试差异,显示振动问题。
685	The results return to the Design Agent for refinement.	结果回到设计代理公司进行完善.
686	The engineer reviews the refined designs	工程师审查精细设计
687	and selects the final one.	并选择最后一个。
688	Different goals, different expertise,	不同的目标,不同的专门知识,
689	the same building blocks,	同样的构件,
690	open models, CUDA-X accelerated tools	开放式模型, CUDA-X 加速工具
691	and the Open Shell Secure Runtime,	和开放的贝壳安全运行时间,
692	brought together with the NVIDIA Agent Toolkit.	与NVIDIA代理工具包合并。
693	Isn't that great?	是不是很棒吗?
694	Some people said that AI is here,	有人说AI在这里
695	it's gonna be the end of software.	这会是软件的终结。
696	It's gonna be end of tools company.	工具公司就要结束了
697	It's exactly the opposite.	则正相反.
698	AI is here, there'll be more users	AI在这里,会有更多的用户
699	of these tools that we created over time than ever before.	我们创造的这些工具
700	Every single tools company will now become an AI company.	现在每个工具公司都会成为AI公司.
701	They'll be an agents company.	他们会是代理公司。
702	Not only will they offer tools,	他们不但提供工具,
703	but they'll offer you specially optimized agents	但他们会提供你特别优化的特工
704	that are incredibly good at using their tools.	他们非常善于使用工具
705	Cadence and CrowdStrike and Dassault, IQVS	卡登斯和人群冲击和达索尔,IQVS
706	and just about every enterprise company out there	大约每个企业公司在那里
707	are all building agents	都是建筑经纪人
708	and we're just delighted to be working with them.	我们很高兴和他们合作
709	Building agents is not that hard.	建筑经纪人没那么难
710	It's got four, five components.	它有四个,五个组件。
711	It's got the model, the harness,	它有模型, 绳子,
712	both of them have closed and open versions,	他们俩都有封闭和开放的版本,
713	but the Open Shell Secure Runtime	但打开贝壳安全运行时间
714	and then most importantly, the knowledge and tools	最重要的是知识和工具
715	and those knowledge and tools belong to you.	那些知识和工具属于你
716	That's your expertise.	这就是你的专长
717	And so all of this is now possible to do	所以现在一切都可以了
718	and I'll show you more in just a second.	我马上就给你看更多
719	But before I do that, let's talk about physical AI.	但在我这样做之前, 让我们谈谈身体AI。
720	That's really the next wave of AI.	这真的是下一轮AI.
721	AI that understands the physical world	了解物理世界的AI
722	can reason about it because the physical world is diverse.	因为物理世界是多种多样的
723	It is complicated.	这很复杂。
724	It's got extremely long tails	尾巴很长
725	and to reason about it and use appropriate tools	并以此为理由,并使用适当的工具;
726	to interact with the physical world.	与物理世界互动。
727	Whenever people think about physical worlds,	每当人们想到物理世界
728	they think robotics.	他们认为机器人。
729	In fact, one of my favorite physical world applications	事实上,我最喜欢的物理世界应用之一
730	is the radio, cell phone, cell towers	是收音机、手机、手机塔
731	and the reason for that is this.	而原因就是这个。
732	It uses instead of cameras,	它用相机代替摄像头
733	instead of speakers and microphones,	而不是扬声器和麦克风,
734	it uses radio waves to sense the world	它用无线电波来感知世界
735	and it reasons about it, it processes information,	它的原因,它处理信息,
736	reasons about it and interacts with the world	与世界互动
737	using beamforming.	使用束状。
738	So in fact, this is an application, a technology	所以事实上,这是一个应用, 一个技术
739	that is just perfectly suited for AI.	这完全适合AI。
740	AI will revolutionize the way we do telecommunications.	AI将改变我们的电信方式。
741	It will enhance our spectrum efficiency tremendously	这将极大地提高我们的频谱效率
742	just as we've done so with ray tracing.	就像我们在追踪射线时所做的那样
743	All of the graphics that you've seen	所有你看到的图形
744	completely 100% ray traced.	完全100%的射线追踪。
745	The amount of computation we thought necessary in the past	我们过去认为必要的计算量
746	and now that we apply AI to is a million to one difference	而现在,我们应用AI 是一百万的区别
747	and so we're gonna bring to the telecommunications network	所以我们要带电信网络
748	an enormous step up in spectrum efficiency	* 频谱效率大幅提高
749	but beyond that, we're also gonna revolutionize	但除此之外,我们还要去革命
750	all of these cell towers.	所有这些细胞塔。
751	There's 100,000 sites around the world,	全世界有十万个网站
752	around the Western world,	在西方世界,
753	about 100 gigawatts of distributed data centers	分布式数据中心约100千兆瓦
754	that will in the future become	未来将变成
755	a distributed AI infrastructure.	a 分布式人工智能基础设施。
756	We're delighted to partner with many companies	我们很高兴和很多公司合作
757	here in Europe and just this last week,	在这里,在欧洲, 就在上周,
758	after working on this for nearly eight years,	经过近八年的努力
759	NVIDIA made the first commercial phone call	NVIDIA打了第一个商业电话
760	与诺基亚的T-Mobile网络
761	and so now CUDA makes phone calls, the traditional way.	因此,现在CUDA 打电话,传统的方式。
762	Then once it can do that,	一旦它能做到这一点,
763	then we can layer AI on top of it	然后我们就可以把人工智能分层了
764	and so the revolution is going to begin.	革命就要开始了
765	This is really exciting opportunity for us.	这对我们来说真是令人兴奋的机会。
766	The other physical AI application	另一个物理AI应用程序
767	that I'm super excited about	我很兴奋
768	is essentially robotic radiotherapy.	本质上是机器人放射疗法。
769	The tumor is changing in shape,	肿瘤在变形,
770	the physical body is changing while you're in therapy.	身体正在改变当你在治疗。
771	The robot, this adaptive radiotherapy system	机器人,这个适应性放射治疗系统
772	has to perceive the changing environment,	必须了解不断变化的环境,
773	understand where the tumor is,	了解肿瘤在哪里,
774	segment it just perfectly	片段完美无缺
775	and direct, precisely direct the radiotherapy accordingly.	并直接,精确地指导放射治疗
776	Essentially robotic, that's right, go ahead.	基本上,机器人,没错,继续。
777	This is a form of robotic surgery.	这是机器人手术的一种形式。
778	Now of course, the highest volume,	当然,最高的音量,
779	the most near term robotics application	最近期的机器人应用
780	or physical AI application that everybody recognized	或每个人都承认的物理AI应用
781	has been something we've been working on	我们一直在研究
782	for over a decade and now we recognize	十多年了,现在我们承认
783	that every single car company will want to make their cars	每个汽车公司都想制造他们的汽车
784	part of the ride hailing fleet.	部分的车驾冰雹舰队。
785	Today we have ride hailing and it's part of Uber	今天我们骑着冰雹,它是Uber的一部分
786	and here in Europe we have Bolt	在欧洲我们有博尔特
787	我们有Grab和Lyft等等
788	and all these different ride hailing services.	和所有这些不同的车欢呼服务。
789	In the future there'll be robo taxis	以后会有计程车
790	that will be connected to those ride hailing services	与那些欢呼服务相连
791	as well but it won't be from just one brand	也好,但不只是一个品牌
792	and so we recognize that building these systems	因此,我们认识到建立这些系统
793	are incredibly expensive,	非常昂贵,
794	particularly the software that goes on top of it.	特别是上面的软件
795	We decided to create one standard architecture,	我们决定建立一个标准建筑,
796	14 different cameras, nine radar systems,	14个不同的摄像机,9个雷达系统,
797	three lidars, umpteen ultrasonics,	三里达,超音速
798	this cocoon of a shell of sensors	这只探测器的茧
799	由双NVIDIA GPU系统供电
800	for the AV computing system.	为AV计算系统.
801	We turned this into an industry standard	我们把这个变成了行业标准
802	and now it's being designed into one car fleet	现在它被设计成一个车队
803	after another.	后别.
804	On top of it are AV software companies	上面还有AV软件公司
805	and we are in fact one of them as well	其实我们也是其中之一
806	to build an autonomous robo taxi capability	建设自动驾驶出租车能力;
807	which then is connected into the ride hailing services.	然后连接到冰雹服务中心
808	This ecosystem is being built out.	这个生态系统正在建设中。
809	The world travels about a trillion miles per year	世界每年旅行一万亿英里
810	around the world.	环绕世界.
811	Many of those miles in the future,	未来许多英里
812	you know a dollar or two per mile	你知道每英里一两块钱
813	are going to be on robo taxis.	将会是 robo出租车。
814	So this is one giant opportunity	因此,这是一个巨大的机会
815	but it's incredibly fragmented	但它令人难以置信的破碎
816	and so creating an industry standard, Hyperion,	于是创造了一个行业标准,Hyperion,
817	and working with an entire ecosystem,	与整个生态系统合作,
818	we could create essentially a giant virtual fleet	我们基本上可以建立一个巨大的虚拟舰队
819	that everybody can connect into.	每个人可以连接。
820	Today, I'm super proud to announce	今天,我非常骄傲地宣布
821	that four more brands, four more cars are going to,	4个品牌,4个汽车要,
822	lines are going to come in.	线路会进来的
823	The VW van, I just can't imagine anything more iconic.	越战车,我无法想象任何更具有标志性的东西。
824	The JLR brand new car, the Volvo fleet	JLR新车沃尔沃车队
825	as well as Mercedes next generation	还有下一代梅赛德斯
826	也会是Hyperion兼容的
827	and be part of this network,	并成为这个网络的一部分,
828	this ride hailing network, robo taxi fleet.	这车的冰雹网络, robo出租车车队。
829	欢迎来到Hyperion
830	The next major industry, the next major industry,	下一个大工业,下一个大工业,
831	the next great industry and this one is likely	下一个伟大的产业和这个有可能
832	to be extremely large is humanoid robotics.	极大的就是人造机器人。
833	There's a reason why robotics is hard to do	机器人很难做到是有原因的
834	because all the different work is rather diverse	因为所有不同的作品都相当不同
835	and so building a special configuration	因此,建立一个特殊配置
836	for each one of the diverse tasks	每个不同的任务
837	segments the robotics industry	机器人工业
838	into all of these different configurations	进入所有这些不同的配置
839	which prevents the technology flywheel to happen.	防止技术飞轮发生
840	The reason why the cell phone has advanced so quickly,	手机这么快推进的原因
841	the reason why personal computers have advanced so quickly,	个人电脑如此快速发展的原因
842	NVIDIA的NVLink 72如此快速推进的原因
843	is because those standardized configurations	这是因为这些标准化配置
844	allows many customers to use it,	允许很多顾客使用它,
845	the volume to therefore be derived,	由此而得之量,
846	that volume drives the technology flywheel	那个音量驱动着技术飞轮
847	and it gets faster, better, incredibly fast.	越快越好越快越好
848	And so we need a robotics configuration	所以我们需要机器人配置
849	that is as standard as possible.	这是尽可能标准。
850	Now human robotics obviously has the ability,	现在人类机器人显然有能力
851	it probably isn't perfect for everything	可能不是完美无缺
852	but it's likely to work for many things.	但它可能为很多事情工作。
853	As a result, this is likely going to be	因此,这可能是
854	the largest volume robotics configuration.	最大的量级机器人配置。
855	Now if we could just create the AI	现在,如果我们可以创建AI
856	that allows it to learn very quickly,	让它能很快学习
857	then for small medium businesses	中小型企业
858	who don't have an entire robotics team	没有一个机器人团队
859	to teach and program the robot	教授和编程机器人
860	can now show the robot what needs to be done.	现在可以告诉机器人需要做什么。
861	The robot because it has general capabilities	机器人因为拥有一般能力
862	and general intelligence	和一般情报
863	would be able to learn and adapt quickly.	能够迅速学习和适应。
864	And so this is likely to be the next major platform,	所以这很可能是下一个主要平台,
865	technology platform.	科技平台.
866	We're working with so many different companies,	我们和很多不同的公司合作
867	you will recognize many of them here in Europe.	在欧洲你会认出他们中的很多人
868	This is your great opportunity.	这是你的大好机会
869	No country, no region has dedicated itself	没有国家,没有地区有奉献
870	to making megatronics, electronics, mechanical systems	制造巨星、电子、机械系统
871	at the level of precision and craft as the European region.	作为欧洲区域的精确度和工艺度。
872	So many companies here are incredibly good at manufacturing.	这里有很多公司都非常擅长制造。
873	This is going to be the next high volume industry.	这将是下一个高额工业。
874	And you could see there's so many companies	你可以看到有很多公司
875	that are now part of it	现在成为它的一部分
876	and we're delighted to be partnered with you.	我们很高兴能跟你合作
877	Now in order to create a physical AI system,	现在为了建立物理AI系统,
878	you need three computers.	你需要三台电脑
879	The first computer is the one we've been talking about	第一台电脑是我们一直在谈论的
880	and everybody knows quite well	人人都很了解这个
881	which is training the AI model.	这是训练AI模式。
882	The second computer is where we have the AI	第二台电脑就是我们有AI的地方
883	learn how to do its skills.	学会如何发挥自己的技能。
884	You can't just have the AI,	你不能只是拥有AI,
885	program the AI and have the AI come out	编程AI 让AI出来
886	into the physical world	进入物理世界
887	and try to do whatever it wants to do.	试着做它想做的事
888	You do it inside a digital twin.	你在数码双胞胎内部做
889	We create a digital twin called Omniverse	我们创造了一个名为Omniverse的数字双胞胎
890	and you've seen us talk about that repeatedly	你见过我们反复谈论
891	and in fact just now,	事实上就在刚才
892	when you saw all the CUDA-X libraries,	当你看到所有的 CUDA -X 库,
893	they're all being done inside a digital twin world.	它们都在数码双胞胎世界里完成
894	That digital twin world is called Omniverse.	那个数码双胞胎世界叫做Omniverse。
895	Omniverse是一个连接的库
896	into all kinds of platforms and tools,	进入各种平台和工具,
897	Siemens and Dassault and Cadence and so on and so forth.	西门子和达索尔,卡登斯等.
898	And so you could see these digital twin systems	这样你就能看到这些数码双子系统
899	being created.	正在创建。
900	The second computer.	第二台电脑
901	The third computer is the robotics computer.	第三台计算机是机器人计算机.
902	That computer integrates in the case of	计算机集成到
903	the self-driving car, the robotaxi,	自驾车机器人轴
904	我们称它为Hyperion。
905	Hyperion由电脑本身组成,
906	the functional safety system,	功能安全系统,
907	the active safety system,	积极的安全系统,
908	an operating system designed for physical AI safety.	用于物理AI安全的操作系统。
909	Well, we need to do exactly the same for robotic systems.	我们需要对机器人系统做同样的事情。
910	So that every single time you start to develop research	所以每次你开始研究
911	and start to work on your own robotic systems,	开始研究你自己的机器人系统
912	you don't have to start from scratch	你不必从头开始
913	to cobble up all of the software and computing systems	将所有软件和计算系统拼接起来
914	and the sensors and all the mechanical systems.	传感器和所有机械系统
915	That alone would take several years.	光这样就需要几年时间。
916	And so we decided to build a reference robot.	所以我们决定建立一个参考机器人。
917	这是NVIDIA AGILE一号搭档
918	与Agile机器人学。
919	This robot is going,	这机器人要飞了
920	this reference robot will be available	此参考机器人将可用
921	in the first quarter of next year	下年第一季度
922	and it will come pre-assembled with computer inside,	里面有电脑,
923	software integrated, and it should be able to	软件集成,它应当能够
924	perform quite well right out of the bat.	表现很好右出蝙蝠。
925	And so this is a great starting point	因此,这是一个伟大的起点
926	and it'll be the flywheel by which we can accelerate	它将会是飞轮我们可以加速
927	development of robotics.	机器人的发展。
928	Well, let me start.	好吧,让我开始。
929	Let me end today's speech from where we started.	请允许我从我们开始的地方结束今天的发言。
930	NVIDIA起源于PC革命.
931	It was in 1993 and the PC revolution	是在1993年和PC革命期间
932	was just about to start.	刚刚开始。
933	Remember, Windows 95 kicked it into Turbo Charge.	记住,Windows 95把它踢进涡轮冲锋机
934	We came since then, invented the GPU,	从那时起,我们发明了GPU
935	programmable shading, real-time ray tracing.	可编程阴影,实时射线跟踪.
936	We added CUDA to the computing model	我们在计算模型中加入了CUDA
937	and helped ignite the modern AI revolution.	并且帮助点燃了现代AI革命.
938	And so four years ago, we started to think,	所以四年前我们开始思考
939	what can we do to reimagine the personal computer	我们能做些什么来重新想象个人电脑
940	for this new era?	对于这个新时代?
941	Where the personal computer is not just your tool,	如果个人电脑不仅仅是你的工具
942	your personal computer tool,	您的个人电脑工具,
943	but your personal agent or your personal assistant.	但你的私人经纪人或私人助理
944	And the way you interact with this computer	你和这台电脑互动的方式
945	would be essentially a companion,	在那日,他确是一个伴侣,
946	somebody who helps you do work.	帮助你工作的人
947	And so this computer needs to have exactly the five things	所以,这台电脑需要准确的五件事
948	that I mentioned today.	我今天提到过
949	It needs to have model.	它需要模型。
950	It could be running locally	可能在本地运行
951	or it could be running in the cloud.	或者它可能在云中运行。
952	It needs a harness.	它需要一把吊带。
953	It needs a secure runtime, open shell,	它需要一个安全的运行时间,打开的壳,
954	and it needs access because it's available on your PC.	它需要访问,因为它在你的电脑上。
955	It needs access to knowledge and tools.	它需要获得知识和工具。
956	And the good news is because NVIDIA has been accelerating	好消息是NVIDIA已经加速了
957	tools for so long.	工具这么久。
958	If we could just create a computer	如果我们能创建一台电脑
959	可以加速 OpenGL 和 DirectX
960	and everything CUDA,	和一切库达,
961	NVIDIA做过的一切
962	would run on this computer.	会用这台电脑运行
963	And so this is it.	这就是它。
964	After four years of partnership with Microsoft,	在和微软合作四年后,
965	我们已经创建了RTX Spark.
966	And this is a brand new line of computers.	这是全新的电脑线路。
967	Literally thousands of years,	数千年,
968	thousands of engineering years went into this.	几千年的工程年进入了这个。
969	We've been secretly working on it for four years,	我们秘密研究了四年
970	created a chip called the N1X	创建一个名为 N1X 的芯片
971	and then we just recently announced it	然后我们刚刚宣布
972	with the entire PC industry joining forces.	与整个PC行业联手.
973	We created desktops, laptops, and workstations.	我们创建了台式计算机、笔记本电脑和工作站。
974	This is one petaflops.	这是一只宠物
975	One petaflops, just to put that in perspective,	一只宠物,只是说到这一点,
976	in the year 2016,	2016年时,
977	I delivered one petaflops	我送了一只燕窝
978	and 128 gigabytes of memory.	还有128千兆字节的记忆
979	One petaflops and 128 gigabytes	一只燕窝和128千兆字节
980	to a small company called OpenAI.	给一家叫OpenAI的小公司
981	This is one petaflops and 128 gigabytes.	这是一只燕窝和128千兆字节。
982	The only difference, of course,	当然,唯一的区别
983	is that I can hold this like this	我可以像这样拿着它
984	and DGX One took two people to carry upstairs.	和总干事一个带两个人上楼
985	这是全新的 RTX Spark PC。
986	And if you would like to have a companion workstation	如果你想有一个伴工站
987	that runs Katia and everything,	经营卡蒂亚和一切
988	自动桌面和Adobe
989	and Cadence and Synopsys,	卡登斯和赛诺斯人,
990	you can always get yourself a DGX station	你总能给自己找个DGX站
991	that runs Windows.	运行窗口。
992	And this is 20 petaflops	这是20个宠物
993	and 754 gigabytes of memory.	还有754千兆字节的记忆
994	And so incredible.	如此不可思议.
995	And of course, there's also the desktop version.	当然,还有桌面版本.
996	All of these are now available	现在所有这一切都可以使用。
997	我认为DGX站与RTX Spark,
998	使用 DGX Spark, DGX 站使用 Windows
999	will be available at the end of this year.	将于今年年底提供。
1000	Now the question is,	现在的问题是,
1001	so far I just talked about the five components of an agent	到目前为止,我刚刚谈到一个特工的五个组成部分
1002	and now the question is, how do you build it?	现在的问题是,你是如何建造的?
1003	How do you build an agent that runs,	如何建立一个代理运行,
1004	let's say on one of those computers	说到其中一台电脑
1005	or runs in a cloud or runs on-prem somewhere?	或者在云中奔跑或者在某个地方奔跑?
1006	Well, the answer is it's really easy	答案是,这很简单
1007	and you do it this way.	你这样做。
1008	The way you build an agent is you ask an agent	你建立特工的方式就是问一个特工
1009	to build it for you.	为你建造它。
1010	So this, this is the prompt	所以,这个是提示
1011	that is going to run the next application.	这将运行下一个应用程序。
1012	And what you're about to see is this.	你将要看到的就是这个。
1013	You're about to see a robotics application.	你即将看到机器人的应用。
1014	And while you're watching this demo,	当你在看这个演示时
1015	just remember there's an agent running	记得有个特工在跑
1016	on this personal computer.	在个人电脑上
1017	This demo runs across three computers.	此演示横跨三台电脑.
1018	The first computer is to train.	第一台电脑是用来训练的.
1019	The second is a digital twin.	第二种是数字双胞胎.
1020	And the third is to put that AI into the computer itself.	第三是把人工智能放进电脑里
1021	Okay, so two very, two ideas.	好吧,所以两个非常,两个想法。
1022	The first idea is agents,	第一个想法是特工
1023	how do you build it and run it	如何建造和运行它
1024	and what you can use it to do.	还有你能用它做什么
1025	And the second is physical AI computing model,	第二个是物理AI计算模型,
1026	which consists of three computers.	由三台计算机组成。
1027	Let's take a look.	让我们来看看。
1028	见到Microduck,拥抱面部机器人。
1029	让我们教它看和走与Nvidia RTX Spark。
1030	I use Hermes as my agent harness,	我以赫尔墨斯为经纪人
1031	Nematron和OpenAI的GPT-6 Astra作为我的模特儿
1032	以及CUDA-X加速工具、FreeCAD、Blender和IsaacSim,
1033	Newton Physics Engine and IsaacLab.	牛顿物理引擎和艾萨克拉布.
1034	我会运行所有我的RTX Spark,
1035	except when I need to burst larger workloads to cloud.	除非我需要把更大的工作量推向云层
1036	First, I describe the course	首先,我描述课程
1037	and what my Microduck needs to learn.	还有我的Microduck需要学的东西
1038	On RTX Spark, Hermes and Nemotron plan the course,	在RTX Spark上,赫尔墨斯和尼莫特龙计划了课程,
1039	listing every object required.	列出每个需要的对象。
1040	Astra takes it from there,	阿斯特拉从那里拿走了
1041	使用 FreeCAD 模拟 3D 对象,
1042	Blender 用于应用材料
1043	和 Omniverse SimReady 库以添加物理属性。
1044	In IsaacSim, Astra composes the scene.	在艾萨克辛(IsaacSim)中,阿斯特拉(Astra)为场景作曲.
1045	Astra钻机 Microduck的关节、引爆器和链接
1046	and simulates how Microduck moves with Newton Physics Engine.	并模拟Microduck如何用牛顿物理引擎移动.
1047	Now, training begins.	现在,训练开始了。
1048	In the cloud, thousands of Microducks train in parallel in IsaacLab.	在云中,以艾萨克拉布的数千辆微电车平行运行.
1049	Every fall, a lesson.	每年秋天,一个教训。
1050	Improving the policy every time.	每次完善政策.
1051	模拟Microduck的摄像机,
1052	teaching it to see.	教其观心.
1053	With perception and physics working together,	有了感知和物理合作,
1054	my Microduck can avoid obstacles, navigate tight spaces	我的Microduck可以避开障碍, 导航紧凑的空间
1055	and adapt to its surroundings.	并适应其周围环境。
1056	我把保险单下载到RTX Spark
1057	and use Nemotron and IsaacSim to evaluate my Microduck	利用尼莫特隆和艾萨克辛来评价我的Microduck
1058	with multi-robot orchestration.	和多机器人的管弦乐
1059	I deploy the validated policy to the real duck	我将经过验证的保单用在了真正的鸭子上
1060	and he's on his way.	他马上就到
1061	NVIDIA RTX Spark. (原始内容存档于2018-07-29).
1062	Build and run AI agents and their accelerated tools	构建和运行人工智能代理及其加速工具
1063	right on your laptop.	在你的笔记本电脑上。
1064	Hey, guys.	嘿,伙计们。
1065	Oh, yeah?	是吗?
1066	Okay, come on.	好吧,来吧。
1067	Here, come over here.	过来,过来
1068	Show people you can walk.	向人们展示你可以走路。
1069	You have a camera.	你有一个相机。
1070	Oh, yeah?	是吗?
1071	Don't trip.	不要绊倒。
1072	Can you guys sit down?	你们能坐下吗?
1073	Okay, stand up.	好吧,站起来。
1074	Take a nap.	睡个觉。
1075	You need to take a nap.	你得睡个觉
1076	Okay, stand up.	好吧,站起来。
1077	Did you know that you were created in here?	你知道你是在这里创造的吗?
1078	你是在RTX Spark上创建的.
1079	I know.	我知道,我记得
1080	And trained you in the cloud	在云中训练你
1081	on a whole bunch of NVIDIA supercomputers.	在一堆NVIDIA超级计算机上
1082	And you learned your skills right here on RTX Spark	你在RTX Spark上学到了自己的技能
1083	in a digital twin, the Omniverse digital twin.	在数字双胞胎中,Omniverse数字双胞胎.
1084	Isn't that incredible?	这不是不可思议吗?
1085	I know.	我知道,我记得
1086	You guys are so cute.	你们真可爱
1087	And your parents are called Pollen.	而父母名为波伦.
1088	P -O -L -L
1089	Uh-huh.	对
1090	Pollen Robotics.	波伦机器人
1091	And apparently, everybody loves you guys so much.	显然,每个人都爱你们这么多。
1092	You're sold out.	你卖光了。
1093	I know, so if you go to the Pollen website, everybody,	我知道,所以如果你去波伦网站,每个人,
1094	everybody watching this,	大家看这个
1095	you know, there's going to be like 20 million viewers,	会有两千万观众
1096	everybody watching this, go to the Pollen website	大家看这个,去波伦的网站
1097	and place an order.	并发布命令。
1098	I think this startup company is like 10 people.	我觉得这家创业公司像10个人
1099	They might blow up.	他们可能会爆炸。
1100	It's going to be so exciting.	一定会很刺激的
1101	Everybody should have one.	每个人都应该有一个。
1102	You guys are so cute.	你们真可爱
1103	And so we'll just have an RTX Spark	所以,我们只需要一个RTX Spark
1104	and we'll teach you new skills.	我们会教你新的技能
1105	Isn't that right?	是吗?
1106	Oh, look at that.	哦,看看那个。
1107	That's incredible.	不可思议.
1108	Can you do a somersault?	你能做点什么吗?
1109	That's so good.	赞曰.
1110	Wow.	珃
1111	Okay.	摆
1112	Well, guess what?	你猜怎么着?
1113	Okay, so today, what did we talk about?	好吧,那么今天,我们谈了什么?
1114	We talked about Europe building its AI economy.	我们谈到了欧洲建设其AI经济.
1115	And it starts with infrastructure.	它从基础设施开始。
1116	And we announced a brand new,	我们宣布了一个全新的,
1117	brand new system called DGX Pod.	全新的系统叫做DGX Pod.
1118	It makes it easy for you to build AI factories.	这让你很容易建造AI工厂。
1119	And these AI factories don't just run models anymore.	而这些AI工厂不再只是经营模特.
1120	Pay attention.	专心点
1121	Stay focused.	专心点
1122	Come here, come here.	过来,过来
1123	Come over here.	过来
1124	Stay focused.	专心点
1125	And these AI factories don't just run,	这些人工智能工厂不只是运行,
1126	there you go, sit down, yeah.	给你,坐下,是的。
1127	They don't just run AI models anymore.	他们不只是运行AI模型了。
1128	They run entire agents.	他们管理着整个特工
1129	And these agents are consistent of five things.	这些特工有五样东西
1130	There's the large language model that does the perception,	有大型语言模型做感知,
1131	reasoning, planning, and acting.	推理、规划和行动。
1132	And it runs in conjunction with a harness	它和绳子一起运行
1133	that turns it essentially exoskeleton.	基本上,它变成了外骨骼。
1134	And once that harness sits inside,	一旦绳子坐在里面,
1135	runs inside a secure runtime we call open shell,	运行在一个安全的运行时间我们称打开壳,
1136	which has just so many people that have supported it.	它有这么多人支持它。
1137	And you give it access to knowledge and tools.	你让它获得知识和工具。
1138	And this agent runs on top of these computers,	这个特工在电脑上运行
1139	but we're also extending into the next generation,	但我们也延续到下一代,
1140	which is where you guys come in, physical AI.	这就是你们进来的地方,身体AI。
1141	So that's it.	故曰此义.
1142	That's what we talked about.	这就是我们谈论的。
1143	Everybody can build agents.	每个人都可以建立特工。
1144	Every company in Europe will become an AI company.	欧洲的每一个公司都会成为AI公司.
1145	There's no question everybody will become an AI company.	毫无疑问每个人都会成为AI公司
1146	And we're going to have a thriving AI economy here in Europe.	我们将在欧洲有一个繁荣的AI经济。
1147	So after 33 years, what an incredible journey.	所以33年后,多么不可思议的旅程。
1148	And we made a really cool video.	我们做了一个很酷的视频。
1149	You guys want to watch it together?	你们想一起看吗?
1150	Okay, come on.	好吧,来吧。
1151	Come over here.	过来
1152	Just come over here just a little bit.	只是过来一点点。
1153	We'll get out of the way.	师便出.
1154	Come on.	来吧。
1155	Top speed.	顶速.
1156	How the heck are we supposed to follow that?	我们该怎么跟他们走?
1157	Well, there is one thing that can make this crowd go even crazier.	有一件事能让人群更加疯狂
1158	What is it?	这是什么?
1159	You won't believe this!	你不会相信的!
1160	GTC has just gone full Eurovision!	GTC刚刚走全欧视!
1161	It begins.	它开始。
1162	Agents working by your side.	你身边的特工
1163	Europe's moment has arrived.	欧洲的时刻到了。
1164	But if you missed what we said at GTC,	但如果你错过我们在GTC说过的话
1165	we'll break it down for you.	我们会帮你破解的
1166	Germany!	德国!
1167	This is a story all about how the world's built.	这是一个关于世界如何建设的故事。
1168	First power plants let energy scale.	第一发电厂允许能源规模。
1169	Then came data centers.	然后是数据中心
1170	Now we've got mail.	现在我们有邮件了
1171	Now we'll be past turning coal into watts.	现在我们可以把煤变成瓦特了
1172	AI factories turn sand into thoughts.	AI工厂将沙子变成思想.
1173	Move over oil to S&P too.	将石油也移到 S&P 。
1174	There's something more investable.	还有更值得投资的
1175	我们的GPU。
1176	It works like an asset.	它的工作就像一个资产。
1177	Reducing green.	减少绿色。
1178	Looks like a chip.	看起来像芯片
1179	Acts like a cash machine.	表现得像个现金机器
1180	They make revenue while you relax.	你放松的时候,它们能赚钱
1181	The cash flows smoother than this sun.	现金流比太阳更顺畅
1182	Agents code and make decisions.	代理密码和决定。
1183	Give them freedom but control their missions.	给他们自由,但控制他们的任务
1184	Open shell on bearers.	打开背包
1185	Checking the list.	检查名单。
1186	No credentials.	无证书。
1187	Access dismissed.	禁止进入
1188	Sentry on blue field.	蓝田哨兵.
1189	Never clocks off.	永远不要关闭时钟。
1190	Protecting the sandbox.	护持沙盒.
1191	Like Hasselhoff.	就像哈塞尔霍夫
1192	We advance AI but limit risks.	我们推进AI,但限制风险。
1193	We're responsibly optimistic.	我们很乐观
1194	RTX Spark使特工安全而明亮.
1195	Our spot rates keep the price just right.	我们的现货价格保持了正确的价格。
1196	We do quantum rexes.	我们做量子量轴。
1197	Gaming all in one.	一起赌博
1198	Cause you can't spell fungible without fung.	因为你不能拼写可替换没有可替换。
1199	Now it's learning the rules of gravity friction.	现在它正在学习重力摩擦的规则.
1200	And using real tools.	并使用真正的工具。
1201	With Isaac Lab and Cosmos.	与艾萨克实验室和宇宙。
1202	They train on their own.	他们自己训练。
1203	It's like college for robots.	这就像大学机器人。
1204	Without student loans.	没有学生贷款。
1205	DSX pocks同步给出AI.
1206	Less power, more brains.	力量少,头脑多.
1207	Like LEGO's that thing.	就像LEGO就是这样
1208	QVS,Neymatron, 和库莫在这里。
1209	Throwing knowledge faster than a Berlin cashier.	比柏林的收银员还快
1210	Your full stack.	你满堆的
1211	It's all on its checklist.	全都在清单上
1212	AI's five layer cake.	AI的五层蛋糕.
1213	It eats it for breakfast.	它吃它作为早餐。
1214	Eurracle系统领先赛事.
1215	It loves open source.	爱开源.
1216	Just ask.	便问.
1217	Hug and faith.	拥抱与信仰.
1218	The continent is rising.	大陆正在崛起。
1219	And the best is yet to come.	而最上者未至.
1220	Come on Europe.	来吧欧洲。
1221	Let's sing as 1TC.	让我们唱为1TC。
1222	Welcome to GTC.	欢迎来到GTC.
1223	I want to thank all of you.	我想感谢你们所有人。
1224	All the sponsors.	所有赞助者.
1225	You can see all the names of companies that are here.	你可以看到这里所有公司的名字.
1226	All the partners we've been working with for 33 years.	我们合作了33年的所有合伙人
1227	Microduck.	微达克.
1228	My new friends.	我的新朋友们
1229	Let's work together to build Europe's AI economy	我们一起建设欧洲的AI经济
1230	and put AI to work.	并让AI工作。
1231	All right.	好吧,我们走。
1232	Have a great GTC.	有一个伟大的GTC。

该视频共有字幕 1240 条。解锁更多字幕为会员功能，请移动到价格

下载安装