(圖說：雪寶登場。圖片來源：NVIDIA GTC 2026 Keynote。)
✳️ 用 Token 經濟思維看 GTC 2026
每年 GTC 結束，社群上都在比規格。哪顆晶片算力多高、NVLink 頻寬多大、Vera Rubin 比 Blackwell 快幾倍。但 Jensen 今年拿出來說「這是我最好的一張投影片」的，不是某顆晶片的規格表，而是一張涵蓋整個結構化資料生態系的架構圖。上面列了 Snowflake、Databricks、Amazon EMR、Google BigQuery 等等 CSP Engines 以及各式各樣的資料儲存方案，最底下是 NVIDIA 的 cuDF 加速引擎。他說團隊每次都勸他「別放這張」，太複雜了，但他堅持要講。核心觀點：「結構化資料是企業運算的 ground truth」，是所有 AI 的地基。
這張圖的重點不在個別平台，而在加速的理由。過去加速結構化資料是為了做更多、更便宜、更頻繁，「還行就好」。但未來 AI 理解資料和使用資料的速度遠超人類，不加速整理就跟不上。Nestle 用 Watson X 加速供應鏈，5 倍快、降本 83%，速度、規模、成本三重好處。NVIDIA 為此建了兩個基礎平台：cuDF 是「資料框架的 RTX」，處理結構化資料；cuVS 處理非結構化資料。後者更關鍵：全球 90% 的資料是非結構化的，過去「對世界完全沒用」，直到 AI 的多模態理解讓它們變得可搜尋。（我們今年第一季陪製造業客戶一起重新探索他們的 ERP 資料，不做不知道，一做下去也是整個嚇爛，通用模型 Opus 4.6 對資料的理解能力讓人瞠目結舌，有興趣的話我們另外約。）
好，假設貴司將資料層加速到位了，下一個問題是：AI 產出的東西怎麼定價？
Jensen 說「token 是新的大宗商品」，所有商品一旦成熟都會分層定價。他秀了一張圖，用吞吐效率和互動速度兩個軸拉出四個 tier：free tier 零成本，medium $3，high $6，premium $45，都以每百萬 token 計價。模型越大、context 越長、速度越快，價格就越高。他還口頭延伸：未來甚至可能出現每百萬 token $150 的等級，對研究團隊來說根本不算什麼。這跟我們熟悉的任何商品市場是類似的邏輯，只是商品從石油、電力，變成了 token。
（關於定價，讓我想到 Werner Vogels 在 AWS re:Invent 2023 年度開發者大會提出的 “The Frugal Architect” 框架。他的 Law 2：「持久的系統會讓成本維度和營收維度對齊。」他舉了手機吃到飽方案的例子，電信商用月費吃到飽定價，但成本按數據量計算。一開始沒問題，直到消費者開始串流 Netflix，數據用量暴增，營收固定，成本隨用量直線上升，定價維度和成本結構脫鉤，商業模式就炸裂了。回到 token 經濟來看：token 的成本已經按量分層了，但許多企業的營收模式還綁在人頭數或月費上。Vogels 說的「對齊」，是讓這兩條軸收斂到同一個維度，否則規模放大的是缺口，不是利潤。）
Jensen 還丟了一個更大的觀點：每家 SaaS 公司都會變成 GaaS 公司，也就是 agentic as a service。他拿 Linux、Kubernetes、HTML 來類比，說每個時代都有一個關鍵基礎設施在對的時間出現，讓整個產業抓住它去做事。過去的企業 IT 是人用工具處理資料和檔案，未來是自主式 AI agent 在執行工作流程，而 agent 消耗的核心資源就是 token。企業評估 AI 基礎設施時，不能只看算力規格，要看整條鏈：底層結構化資料的處理效率、中間 token 的成本結構、上層服務的定價模式，三者對齊才撐得住規模化。（McKinsey 和 Accenture 報告提到 90% 大型組織計劃增加 AI 投資，但只有約 20% 重新設計了工作流程。沒對齊就狂買 GPU 灌 token，跟買了健身房會員卡但從不去差不多意思。但預算還是要先過啦！沒過預算還在那邊問東問西是來吃什麼霸王餐！）
所以下次看到 AI 基礎設施的消息，可以試著問自己幾件事。我的資料層準備好被 agent 高頻存取了嗎？我的 token 成本結構和營收模式是對齊的，還是各走各的？如果每家 SaaS 都在變成 GaaS，我的服務在結構化資料地基、token 經濟效率、成本對齊這三個軸上各站在什麼位置？想清楚這幾件事，比追規格表跑要實際得多。（先從盤點資料、盤點流程這些基本功開始做起，基本上該焦慮的不是在 AI 時代有這麼多工具該學哪一個，而是連基本功都不願意扎實打底，貪求速食與囫圇吞棗，噎到可能也是剛好而已？）（很好，今天也食用了誠實豆沙）
✳️ 知識圖譜
(更多關於知識圖譜…)
Token 經濟分層模型
硬體架構演進時間軸
自主式 AI 軟體堆疊
✳️ 延伸閱讀
✳️ 完整筆記
開場
演出前音樂
- This is how intelligence is made.
這就是智慧誕生的方式。 - A new kind of factory.
一種全新的工廠。 - Generator of tokens.
Token 的生成器。
開場影片：Tokens
Tokens：AI 的基本組成元素
- The building blocks of AI.
AI 的基本組成元素。 - Tokens have opened a new frontier.
Token 開啟了一個全新的疆界。 - Turning data into knowledge.
將資料轉化為知識。 - And drawing on all we have learned.
並汲取我們所學的一切。 - Tokens are harnessing a new wave of clean energy.
Token 正在駕馭一波全新的潔淨能源浪潮。 - And unlocking the secrets of the stars.
並揭開星辰的奧秘。 - In virtual worlds, they help robots learn.
在虛擬世界中，它們幫助機器人學習。 - And in the physical world, perfect.
而在真實世界中，則臻於完美。 - Forging new paths.
開闢新的道路。 - And clearing the way for a bountiful harvest.
並為豐收掃清障礙。 - In the moments that matter, tokens are already there.
在關鍵時刻，token 早已就位。
Tokens 的影響力與潛力
- And in the miles between, they never stop.
在漫長的路途之間，它們從未停歇。 - They work where human hands cannot.
它們在人手無法觸及之處運作。 - So we may all breathe easier.
讓我們都能呼吸得更輕鬆。 - And the smallest hearts beat stronger.
讓最小的心臟跳動得更有力。 - Tokens are helping us break new ground.
Token 正在幫助我們開拓新天地。 - On a scale never attempted.
以前所未有的規模。 - To empower the world.
賦能整個世界。
歡迎致詞與大會概覽
- So we can reach One separation confirmed
讓我們能夠抵達，確認了一個分界點。 - Well beyond it We take the next great leap Into a bright new future Built for all mankind
遠遠超越它，我們邁出下一個偉大的飛躍，走進為全人類打造的光明新未來。 - Is where it all begins Welcome to the stage NVIDIA founder and CEO
這就是一切開始的地方，歡迎 NVIDIA 創辦人暨執行長上台。 - Welcome to GTC I just want to remind you This is a tech conference
歡迎來到 GTC，我只想提醒大家，這是一場科技大會。 - All these people lining up So early in the morning All of you in here It’s great to see you
所有人一大早就來排隊，你們都在這裡，真高興見到大家。 - GTC GTC We’re going to talk about technology We’re going to talk about platforms
GTC，GTC，我們將會談論技術，也會談論平台。 - NVIDIA has three You think that we Mostly talk about one of them It’s related to CUDA X
NVIDIA 有三個平台，你們以為我們大多只談其中一個，那就是跟 CUDA X 相關的平台。 - Our systems is another platform And now we have a new platform Factories
我們的系統是另一個平台，而現在我們有了一個新平台：工廠。 - We’re going to talk about all of them And most importantly we’re going to talk about ecosystems
我們將會談論所有平台，而最重要的是，我們將會談論生態系。 - But before I start talking about ecosystems Let me thank our pre-game show hosts
但在我開始談論生態系之前，讓我先感謝我們的暖場節目主持人。 - I thought they did a great job Capital NVIDIA’s first venture capitalist Baker NVIDIA’s first major institutional investor
我覺得他們做得非常好。Capital 是 NVIDIA 的第一位創投投資人，Baker 是 NVIDIA 的第一位主要機構投資人。 - These three people are deep in technology Deep in what’s going on
這三位都深耕科技，深入了解當前的發展趨勢。 - And of course They have just a really broad reach Of technology ecosystem
當然，他們在科技生態系中也有非常廣泛的影響力。 - And then of course All of the VIPs that I hand selected To join us today All-star team
然後當然還有所有我親自挑選的 VIP 嘉賓，今天加入我們的全明星陣容。 - I want to thank all of you for that
我要感謝你們所有人。 - I also want to thank all the companies That are here
我也要感謝所有在場的企業。 - NVIDIA as you know Is a platform company We have technology We have our platforms We have a rich ecosystem
如大家所知，NVIDIA 是一家平台公司，我們有技術、有平台、有豐富的生態系。 - There are probably 100%
這裡大概有百分之百。
AI 層層蛋糕與應用
- trillion dollars of industry Here 450 companies sponsored this event I want to thank you
數兆美元規模的產業在這裡，450 家企業贊助了這次活動，我要感謝你們。 - 1,000 Technical sessions 2,000 speakers
1,000 場技術議程，2,000 位講者。 - This conference is going to cover Every single layer Of the five layer cake of artificial intelligence
這場大會將會涵蓋人工智慧五層蛋糕的每一個層次。 - From land power and shell To infrastructure To the platforms
從土地、電力與外殼，到基礎建設，再到平台。 - And of course the most important And ultimately What’s going to get this industry taken off Is all of the applications
當然最重要的，也是最終會讓這個產業起飛的，就是所有的應用。 - What it all began It all began here This is the 20th
一切從哪裡開始的？一切都從這裡開始。這是第 20 屆。
1️⃣ CUDA 平台：二十週年
CUDA 的二十年旅程與生態系
- anniversary of CUDA We’ve been working on CUDA for 20 years
CUDA 週年紀念，我們已經在 CUDA 上投入了 20 年。 - For 20 years we’ve been dedicated to this architecture
20 年來我們一直致力於這個架構。 - This revolutionary invention CIMD Single instruction Multi-threaded
這項革命性的發明 CIMD，單指令多執行緒。 - Writing scalar code Could spawn off into Multi-threaded application Much much easier to program
撰寫純量程式碼就能衍生出多執行緒應用程式，程式設計變得容易許多。 - We recently added tiles So that we could help people program Tensor cores
我們最近加入了 tiles，好讓大家能夠對 Tensor cores 進行程式設計。 - And these structures of mathematics That are so foundational To artificial intelligence today
以及這些數學結構，它們是當今人工智慧的基礎。 - Thousands of people have been working on this for decades
數千人已經為此投入了數十年。 - Thousands of tools And compilers And frameworks And libraries
數千種工具、編譯器、框架和函式庫。 - In open source There’s a couple of hundred thousand Public projects
在開源領域中有數十萬個公開專案。 - CUDA literally is integrated Into every single ecosystem
CUDA 實際上已經整合進每一個生態系中。 - Basically describes A hundred percent of NVIDIA’s strategies
基本上描述了 NVIDIA 百分之百的策略。 - You’ve been watching me talk about this slide From the very beginning
你們從一開始就看著我在講這張投影片。 - And ultimately The single hardest thing to achieve Is the thing on the bottom Installed base
而最終，最難達成的就是底部那個東西：裝機量。 - It has taken us 20 years To now have built up Hundreds of millions of GPUs And computing systems around the world That run CUDA
我們花了 20 年，如今已在全球建立了數億個執行 CUDA 的 GPU 和運算系統。 - We are in every cloud We’re in every computer company We serve just about every single industry
我們存在於每一個雲端、每一家電腦公司，服務幾乎每一個產業。 - The installed base Is the reason why The flywheel is accelerating
裝機量正是飛輪加速的原因。 - The installed base is what attracts Developers to the cloud Who then creates new algorithms That achieves a breakthrough
裝機量吸引開發者進入雲端，他們隨後建立新的演算法並實現突破。 - For example Deep learning There are so many others
例如深度學習，還有許多其他的。 - Those breakthroughs Leads to entirely new markets Which builds new ecosystems around them
這些突破引領了全新的市場，並在其周圍建立新的生態系。 - With other companies that join Which creates a larger installed base
隨著其他公司加入，建立了更大的裝機量。 - This flywheel This flywheel Is now accelerating
這個飛輪，這個飛輪，現在正在加速。 - The number of downloads of NVIDIA libraries Is incredibly accelerating
NVIDIA 函式庫的下載數量正在驚人地加速成長。 - It’s at a very large scale And growing faster than ever
規模非常龐大，而且成長速度前所未有。 - This flywheel Is what makes This computing platform Able to sustain So much applications So many new breakthroughs
這個飛輪使得這個運算平台能夠支撐如此多的應用和如此多的新突破。 - But most importantly It also enables These infrastructures To have extraordinarily useful life
但最重要的是，它還使這些基礎設施擁有極長的有效使用壽命。
飛輪效應與平台策略
- And the reason for
原因是。 - Is very obvious There are so many applications That you can run on NVIDIA CUDA
非常明顯，有太多應用程式可以在 NVIDIA CUDA 上執行。 - We support the entire Every single phase of the AI life cycle
我們支援 AI 生命週期的每一個階段。 - We address every single phase of the AI life cycle
我們涵蓋 AI 生命週期的每一個階段。 - We address every single data processing platform
我們涵蓋每一個資料處理平台。 - We accelerate scientific principle solvers Of all different kinds
我們加速各種不同類型的科學原理求解器。 - And so the application reach Is so great That once you install NVIDIA GPUs The useful life of it Is incredibly high
因此應用的覆蓋範圍非常廣，一旦安裝了 NVIDIA GPU，它的有效使用壽命就非常長。 - It is also one of the reasons why Ampere that we shipped them six years ago The pricing of Ampere in the cloud Is going up
這也是為什麼我們六年前出貨的 Ampere，其在雲端的定價反而在上漲。 - And so all of that is made possible Fundamentally because the Install base is high The flywheel is high The developer reach is great
所有這一切之所以可能，根本原因在於裝機量很高、飛輪效應很強、開發者的觸及範圍很廣。 - And when all of that happens And we continuously update our software
當所有這些同時發生，而我們持續更新軟體。 - The computing cost The combination of accelerated computing Speeding up applications tremendously
運算成本，加速運算的結合，大幅加速應用程式。 - As we continue to nurture And continue to update software Over its life
隨著我們持續維護並在整個生命週期中不斷更新軟體。 - Not only do you get the first time pop get the continuous cost reduction Of accelerated computing
你不僅獲得首次的效能提升，還能持續獲得加速運算帶來的成本降低。 - And we are willing to nurture Willing to support Every single one of these GPUs In the world
我們願意維護、願意支援全世界每一個 GPU。 - Because they are all architecturally compatible
因為它們在架構上全部相容。 - We are willing to do so Because the install base is so large
我們願意這樣做，因為裝機量非常龐大。 - If we release a new optimization It benefits millions
如果我們發布一個新的最佳化，數百萬人都能受益。 - This applies to everybody in the world
這適用於全世界的每一個人。 - This combination of dynamics Is what makes the NVIDIA architecture Expand its reach Accelerating its growth
這種動態的組合使得 NVIDIA 架構擴大了覆蓋範圍，加速了成長。 - At the same time down computing cost Which ultimately Encourages new growth
同時降低了運算成本，最終促進新的成長。 - So CUDA is at the center of it
所以 CUDA 是這一切的核心。 - But our journey to CUDA Actually started 25 years ago GeForce
但我們通往 CUDA 的旅程其實始於 25 年前的 GeForce。 - I know how many of you grew up with GeForce
我知道你們有多少人是伴隨 GeForce 長大的。 - GeForce is NVIDIA’s Greatest marketing campaign And it’s the biggest marketing campaign in the world.
GeForce 是 NVIDIA 最偉大的行銷活動，也是全世界最大的行銷活動。
GeForce 的家與 CUDA 的基礎
- relax This is the house that G-Force made. 25 years ago, we started our journey which led to CUDA. 25 years ago, we invented the programmable shader.
放輕鬆，這就是 GeForce 建造的家。25 年前，我們開始了通往 CUDA 的旅程。25 年前，我們發明了可程式化著色器。 - A perfectly unobvious invention to make an accelerator programmable.
一個完全不顯而易見的發明，讓加速器變得可程式化。 - The world’s first programmable accelerator, the pixel shader. 25 years ago, that led us to explore further and further, 20 years later, 5 years later, the invention of CUDA.
全球第一個可程式化加速器，像素著色器。25 年前，那引導我們不斷探索，20 年後，也就是 5 年後，CUDA 誕生了。 - One of the biggest investments that we made, and we couldn’t afford it at the time, and it consumed the vast majority of our company’s profits, was to take CUDA on the backs of G-Force to every single computer.
我們做過最大的投資之一，當時我們其實負擔不起，它消耗了公司絕大部分的利潤，就是透過 GeForce 將 CUDA 帶到每一台電腦上。 - We dedicated ourselves to create this platform because we felt so strongly about it.
我們全心投入打造這個平台，因為我們對它有強烈的信念。 - We felt so strongly about its potential.
我們對它的潛力深信不疑。 - But ultimately, the company’s dedication to it, despite the hardships in the beginning, believing in every single day, for 13 generations or 20 years, we now have CUDA installed everywhere.
但最終，公司對此的堅持，儘管初期困難重重，每一天都在堅持信念，經歷了 13 個世代或 20 年，我們現在已經讓 CUDA 安裝到了每一個地方。
圖形技術的演進與 AI 的影響
- The pixel shader led to, of course, the revolution of G-Force.
像素著色器當然帶來了 GeForce 的革命。 - And then 10 years ago, we introduced, about 10 years ago, what is it, 8 years ago?
然後大約 10 年前，我們推出了，大概 10 年前，還是 8 年前？ - We introduced RTX.
我們推出了 RTX。 - A complete redesign of our architecture for the modern era of computer graphics.
針對現代電腦圖形時代的一次完整架構重新設計。 - G-Force brought CUDA to the world.
GeForce 將 CUDA 帶給了全世界。 - G-Force, therefore, enabled Alex Krushevsky and Ilya Suskovor and Jeff Hinton, Andrew Ang, and so many others to discover that the GPU could be their friend in accelerating deep learning.
因此 GeForce 讓 Alex Krizhevsky、Ilya Sutskever、Jeff Hinton、Andrew Ng 和許多其他人發現 GPU 可以成為他們加速深度學習的好夥伴。 - It started the big bang of AI. 10 years ago, we decided that we would fuse 10 years ago, we decided that we would fuse programmable shading and introduce two new ideas.
這開啟了 AI 的大爆炸。10 年前，我們決定融合可程式化著色，並引入兩個新概念。 - Ray tracing, hardware ray tracing, which is incredibly hard to do, and a new idea at the time.
光線追蹤，硬體光線追蹤，這極其困難，在當時是一個全新的想法。 - Imagine, about 10 years ago, we thought that AI would revolutionize computer graphics.
想像一下，大約 10 年前，我們就認為 AI 將徹底改變電腦圖形學。 - Just as G-Force brought AI to the world, AI is now going to go back and revolutionize how computer graphics is done all together.
正如 GeForce 將 AI 帶給了全世界，AI 現在將反過來徹底革新整個電腦圖形的運作方式。
神經渲染技術介紹
- Well, today, I’m going to show you something of the future.
好的，今天我要向你們展示一些來自未來的東西。 - This is our next generation of graphics technology.
這是我們的下一代圖形技術。 - We call it neuro-rendering.
我們稱之為神經渲染（neuro-rendering）。
神經渲染視覺化
- The fusion, the fusion of 3D graphics and artificial intelligence.
融合，3D 圖形與人工智慧的融合。 - This is DLSS 5.
這是 DLSS 5。 - Take a look at it.
看看吧。 - Enter whatever comes to your mind as shown to it.
輸入你腦海中浮現的任何東西。 - Hey look at that.
嘿，看看那個。 - We also have our expert VR.
我們還有我們的專業 VR。 - We’ll learn today how it all works.
今天我們會了解這一切是如何運作的。 - Go to our project, Is that incredible?
看看我們的專案，是不是很驚人？ - Computer graphics comes to life.
電腦圖形栩栩如生。
融合概念與可信賴 AI
- Now, what did we do?
那麼，我們做了什麼？ - We fused controllable 3D graphics, the ground truth of virtual worlds, the structured data.
我們融合了可控制的 3D 圖形，虛擬世界的真實基準，也就是結構化資料。 - Remember this word, the structured data of virtual worlds, of generated worlds.
記住這個詞，虛擬世界、生成世界的結構化資料。 - We combined 3D graphics, structured data, with generative AI, probabilistic computing.
我們將 3D 圖形、結構化資料，與生成式 AI、機率運算結合在一起。 - One of them is completely predictive.
其中一個是完全可預測的。 - The other one, probabilistic, yet highly realistic.
另一個是機率性的，但高度逼真。 - We combined these two ideas, combined these two ideas, controlled through structured data, controlled perfectly, and yet generating at the same time.
我們結合了這兩個概念，透過結構化資料進行控制，完美地控制，同時又在生成。 - And as a result, the content is beautiful, amazing, as well as controllable.
結果就是，內容既美麗、驚人，又可控制。 - This concept of fusing structured information and generative AI will repeat itself in the future.
這種融合結構化資訊與生成式 AI 的概念將在未來不斷重現。 - In one industry after another industry after another industry.
在一個接一個的產業中。 - Structured data is the foundation of trustworthy AI.
結構化資料是可信賴 AI 的基礎。
結構化與非結構化資料平台
「最佳投影片」與結構化資料平台
- Well, this is going to scare you a little bit.
好吧，這個可能會嚇到你們一點。 - I’m going to flip the slide, and don’t gasp.
我要翻到這張投影片了，不要倒抽一口氣。 - So we’re going to go through this schematic for the rest of the time.
接下來的時間我們會逐步講解這張架構圖。 - This is my best slide.
這是我最好的一張投影片。 - Every time I ask the team, what’s my best slide?
每次我問團隊，我最好的投影片是哪張？ - Repeatedly, this was it.
反覆地，就是這張。 - They say, don’t do it, Jensen.
他們說，Jensen 不要用這張。 - Don’t do it.
不要用。 - I said, no.
我說，不行。 - These seats are free for some of you.
你們有些人的座位是免費的。 - So this is your price of admission.
所以這就是你們的入場代價。 - So this is structured data.
所以這就是結構化資料。 - You’ve heard of it.
你們聽過它。 - SQL, Spark, Pandas, Velox, some of these really, really important, very large platforms.
SQL、Spark、Pandas、Velox，其中一些是非常重要、規模龐大的平台。
資料框架與企業運算
- Snowboard.
Snowboard。 - Snowflake.
Snowflake。 - Databricks.
Databricks。 - EMR.
EMR。 - Amazon EMR.
Amazon EMR。 - Azure Fabric.
Azure Fabric。 - Google Cloud.
Google Cloud。 - BigQuery.
BigQuery。 - All of these platforms are processing data frames.
所有這些平台都在處理資料框架。 - These data frames are giant spreadsheets, and they hold all of life’s information.
這些資料框架是巨大的試算表，它們承載著生活中所有的資訊。 - This is the structured data, the ground truth of business.
這就是結構化資料，商業的真實基準。 - This is the ground truth of enterprise computing.
這就是企業運算的真實基準。 - Well.
好。
未來 AI Agent 與資料庫使用
- Now we’re going to have AI use structured data.
現在我們要讓 AI 使用結構化資料。 - And we better accelerate the living daylights out of it.
我們最好全力加速它。 - It used to be OK.
以前還可以。 - And we would, of course, we would accelerate structured data so that we could do more.
我們當然會加速結構化資料的處理，這樣我們就能做更多。 - We could do it more cheaply.
我們可以更便宜地完成。 - We could do it more frequently per day and keep the company running at a much more synchronized way.
我們可以每天更頻繁地處理，讓公司以更同步的方式營運。 - However, in the future, what’s going to happen is these data structures are going to be used by AI.
然而，在未來，這些資料結構將被 AI 使用。 - And AI is going to be much, much faster than us.
而 AI 將比我們快得多。 - Future agents are going to use structured databases as well.
未來的 Agent 也會使用結構化資料庫。
非結構化資料挑戰與 AI 解決方案
- And then, of course, the unstructured database, the generative database.
然後，當然還有非結構化資料庫，生成式資料庫。 - This database represents the vast majority of the world.
這個資料庫代表了世界上的絕大部分。 - Vector databases, unstructured data, PDFs, videos, speeches, all of the world’s information, about 90% of what’s generated every single year is unstructured data.
向量資料庫、非結構化資料、PDF、影片、語音，世界上所有的資訊，每年產生的資料約有 90% 是非結構化資料。 - Until now, this data has been completely useless to the world.
直到現在，這些資料對世界來說一直是完全無用的。 - We read it.
我們讀了它。 - We put it into our file system.
我們把它放進我們的檔案系統。 - And that’s it.
就這樣。 - Unfortunately, we can’t query it.
不幸的是，我們無法查詢它。 - We can’t search for it.
我們無法搜尋它。 - It’s hard to do that.
這很難做到。 - And the reason for that is because there’s no easy indexing of unstructured data.
原因在於非結構化資料沒有簡單的索引方式。 - You have to understand its meaning, its purpose.
你必須理解它的含義、它的目的。 - And so now we have AI do that.
所以現在我們讓 AI 來做這件事。 - Just as AI was able to solve multimodality perception and understanding, you can use that same technology, multimodality perception and understanding, to go read a PDF.
就像 AI 能夠解決多模態感知與理解一樣，你可以使用同樣的技術，多模態感知與理解，去閱讀一份 PDF。 - To understand its meaning.
去理解它的含義。 - And from that meaning, embed it into a larger structure that we can search into, we can query into.
然後從那個含義出發，將它嵌入到一個更大的結構中，讓我們可以搜尋、可以查詢。 - NVIDIA created two foundational libraries.
NVIDIA 建立了兩個基礎函式庫。
QDF 與 QVS 平台
- Just like we created RTX for 3D graphics, we created QDF for data frames, structured data.
就像我們為 3D 圖形建立了 RTX 一樣，我們為資料框架、結構化資料建立了 QDF。 - We created QVS for vector stores, semantic data, unstructured data, AI data.
我們為向量儲存、語意資料、非結構化資料、AI 資料建立了 QVS。 - These two platforms are going to be the best.
這兩個平台將會是最好的。 - And we’re going to be the best.
我們將會是最好的。 - We’re going to be two of the most important platforms in the future.
它們將成為未來最重要的兩個平台。 - Super excited to see its adoption throughout the network, this complicated network of the world’s data processing systems.
非常興奮能看到它在整個網路中被採用，這個複雜的全球資料處理系統網路。 - And the reason for that is because data processing has been around a long time.
原因在於資料處理已經存在很長時間了。 - And therefore, so many different companies and platforms and services.
因此有非常多不同的公司、平台和服務。 - It has taken us a long time to integrate deeply into this ecosystem.
我們花了很長時間才深度整合進這個生態系。 - I’m super proud of the work that we’re doing here.
我對我們在這裡所做的工作感到非常自豪。 - And then today, we’re announcing several of them.
今天，我們要宣布其中幾項。 - IBM, the inventor of SQL, one of the most important domain-specific languages of all time, is accelerating Watson X data with QDF.
IBM，SQL 的發明者，有史以來最重要的領域專用語言之一，正在使用 QDF 加速 Watson X data。
加速運算合作夥伴
歷史脈絡與 AI 時代
- Let’s take a look at it.
讓我們來看看。 - Sixty years ago, IBM introduced the System 360, the first modern platform for general-purpose computing, launching the computing era.
60 年前，IBM 推出了 System 360，第一個現代通用運算平台，開啟了運算時代。 - Then SQL.
然後是 SQL。 - This is the system that is used to merge data into a system that uses a declarative language to query data without requiring the computer to be instructed step by step.
這是一個使用宣告式語言來查詢資料的系統，不需要逐步指示電腦。 - And the data warehouse.
以及資料倉儲。 - Each, the foundations of modern enterprise computing.
每一個都是現代企業運算的基礎。 - Today, IBM and NVIDIA are reinventing data processing for the era of AI by accelerating IBM Watson X.data SQL engines with NVIDIA GPU computing libraries.
今天，IBM 和 NVIDIA 正在透過使用 NVIDIA GPU 運算函式庫加速 IBM Watson X.data SQL 引擎，為 AI 時代重新發明資料處理。
AI 的加速運算
- AI needs rapid access to massive datasets.
AI 需要快速存取大量資料集。 - Today’s CPU data processing systems can’t keep up.
當今的 CPU 資料處理系統跟不上。 - Nestle makes thousands of supply chain decisions every day.
Nestle 每天做出數千個供應鏈決策。 - Their order-to-cash data mart aggregates every supply, order, and delivery event across global operations in 185 countries.
他們的訂單到現金資料市集彙整了橫跨 185 個國家全球營運的每一筆供應、訂單和配送事件。 - On CPUs, Nestle refreshed the data mart a few times a day.
在 CPU 上，Nestle 每天只能更新資料市集幾次。 - With accelerated Watson X.data running on NVIDIA GPUs, Nestle can run the same workload five times faster at 83% lower cost.
透過在 NVIDIA GPU 上執行加速版 Watson X.data，Nestle 可以以快 5 倍的速度執行相同的工作負載，成本降低 83%。 - The next computing platform has arrived: accelerated computing for the era of AI.
下一個運算平台已經到來：AI 時代的加速運算。
Dell AI 資料平台
- NVIDIA accelerates data processing in the cloud.
NVIDIA 加速雲端的資料處理。 - We also accelerate data processing on-prem.
我們也加速地端的資料處理。 - As you know, Dell is the world-leading computer systems maker, and they also are one of the world’s leading storage providers.
如你們所知，Dell 是全球領先的電腦系統製造商，同時也是全球領先的儲存供應商之一。 - And they worked with us to create the Dell AI data platform that integrates QDF and QVS to create an accelerated data platform, well, for the era of AI.
他們與我們合作打造了 Dell AI 資料平台，整合 QDF 和 QVS 來建立加速資料平台，為 AI 時代而生。 - And this is an example of what they did with NTT data.
這是他們與 NTT data 合作的案例。 - Huge speed-up.
巨幅加速。
Google Cloud 與成本效益
- This is cloud, Google Cloud.
這是雲端，Google Cloud。 - As you know, we’ve been working with Google Cloud for a very long time.
如你們所知，我們與 Google Cloud 合作已經很長時間了。 - We accelerate Google’s Vertex AI.
我們加速 Google 的 Vertex AI。 - We now accelerate BigQuery, really important framework and really important platform.
我們現在加速 BigQuery，非常重要的框架和非常重要的平台。 - And this is an example of our work together with Snapchat, where we reduced their cost of computing by nearly 80%.
這是我們與 Snapchat 合作的案例，我們將他們的運算成本降低了近 80%。 - When you accelerate data processing, when you accelerate computing, you get the benefit of speed, you get the benefit of scale.
當你加速資料處理、加速運算時，你會獲得速度的好處，也會獲得規模的好處。 - But most importantly, you also get the benefit of cost.
但最重要的是，你還會獲得成本的好處。 - And so all of those come together as one.
所以這一切合而為一。
超越摩爾定律：加速運算
- It was originally called Moore’s Law.
它原本被稱為 Moore’s Law（摩爾定律）。 - Moore’s Law was about getting performance doubling every couple of years.
Moore’s Law 是關於每隔幾年效能就翻倍。 - It’s another way of saying, so long as the price remains about the same, and most computers remained about the same, you’re also getting twice the performance every year.
換一種說法，只要價格維持不變，而大多數電腦也保持不變，你每年也能獲得兩倍的效能。 - Or you’re reducing the cost of computing every single year.
或者說你每年都在降低運算成本。 - Well, Moore’s Law has run out of style.
好吧，Moore’s Law 已經過時了。 - It’s not a new thing.
這不是新鮮事。 - And that’s why we need a new approach.
這就是為什麼我們需要一種新方法。 - Accelerated computing allows us to take these giant leaps forward.
加速運算讓我們能夠實現這些巨大的飛躍。
演算法最佳化與成本降低
- And as you will see later, because we continue to optimize the algorithms, and NVIDIA is an algorithm company, as we continue to optimize the algorithms.
正如你們稍後會看到的，因為我們持續最佳化演算法，而 NVIDIA 是一家演算法公司，我們持續最佳化演算法。 - And because our reach is so large and our install base is so large, we can reduce the computing cost, increasing the performance of our software.
因為我們的覆蓋範圍如此之大、裝機量如此之大，我們可以降低運算成本，提升我們軟體的效能。 - You can see this pattern I just mentioned.
你們可以看到我剛才提到的這個模式。 - I just wanted to show you three versions of it.
我只是想展示三個版本給你們看。 - NVIDIA built the accelerated computing platform.
NVIDIA 打造了加速運算平台。 - It has a bunch of libraries on top.
上面有一堆函式庫。 - I gave you three examples.
我給了你們三個例子。 - RTX is one of them.
RTX 是其中之一。 - QDF is another.
QDF 是另一個。 - QVS.
QVS。 - And we’ll show you a few more.
我們還會展示更多。 - These libraries sit on top of our platform.
這些函式庫建立在我們的平台之上。 - But ultimately, we integrate into the world’s cloud services, into the world’s OEMs.
但最終，我們整合進全球的雲端服務、全球的 OEM 廠商中。 - Together, and other platforms that I’ll show you, together we’re able to reach the world.
共同合作，加上我接下來要展示的其他平台，我們一起能夠觸及全世界。 - This pattern, NVIDIA, Google Cloud, Snapchat, will repeat over and over again and kind of looks like this.
這個模式，NVIDIA、Google Cloud、Snapchat，將會一再重複，大概看起來像這樣。 - And so this is one example, NVIDIA with Google Cloud.
所以這是一個例子，NVIDIA 與 Google Cloud。 - We accelerated Vertex AI.
我們加速了 Vertex AI。 - We accelerated BigQuery.
我們加速了 BigQuery。 - I’m super proud of the work that we’ve done with JAX and XLA.
我對我們在 JAX 和 XLA 方面所做的工作感到非常自豪。 - We are incredible on PyTorch.
我們在 PyTorch 上表現非常出色。 - We’re the only accelerator in the world that’s incredible on PyTorch.
我們是全世界唯一在 PyTorch 上表現出色的加速器。 - We’re the only accelerator in the world that’s incredible on PyTorch and incredible on JAX and XLA.
我們是全世界唯一在 PyTorch 上表現出色，同時在 JAX 和 XLA 上也表現出色的加速器。 - And the customers that we support, the Baseten, the CrowdStrikes, Puma, Salesforce, they’re not our customers, but they’re customers, developers of ours.
我們支援的客戶，Baseten、CrowdStrike、Puma、Salesforce，他們不是我們的客戶，但他們是我們的客戶，我們的開發者。 - We’ve integrated the NVIDIA technologies into that we can then land on the clouds.
我們整合了 NVIDIA 技術，使其能夠落地到雲端。 - Our relationship with cloud service providers are essentially us bringing customers to them.
我們與雲端服務供應商的關係，本質上是我們幫他們帶來客戶。 - We integrate our libraries.
我們整合我們的函式庫。 - We accelerate workloads.
我們加速工作負載。 - And we land those customers in the clouds.
然後我們將那些客戶帶到雲端。 - And so, as you can see, most of our cloud service providers love working with us.
所以，如你們所見，大多數雲端服務供應商都喜歡與我們合作。 - And they’re always asking us to land the next customer on their cloud.
他們總是要求我們將下一個客戶帶到他們的雲端上。
客戶落地與雲端成長
- And I just want to let you know, there are a lot of customers.
我只是想讓你們知道，有非常多的客戶。 - We’re going to accelerate everybody.
我們會加速每一個人。 - And so there will be lots and lots of customers who will be able to land in your cloud.
所以會有非常多的客戶能夠落地到你們的雲端上。 - Just be patient with us.
請對我們有耐心。（Ernest 於弦外之超譯：除了我們，其他人應該做不到（吧）。別急，等我們一下下，請對我們有耐心。）
Google Cloud 與 AWS
- And so this is Google Cloud.
所以這是 Google Cloud。 - This is AWS.
這是 AWS。（Ernest 超譯：你各位知道上菜會有順序，開胃菜喚起味蕾，然後誰才是主菜應該越發明顯了。剩下其他的，就是剩下其他的。） - We’ve been working with AWS a long time.
我們與 AWS 合作很長時間了。 - And one of the areas, one of the things I’m super excited about this year is we’re going to bring OpenAI to AWS.
今年我最興奮的事情之一就是我們要將 OpenAI 帶到 AWS。 - And so it’s going to drive enormous consumption of cloud computing at AWS.
這將會帶動 AWS 雲端運算的巨大消耗量。 - It’s going to expand the reach and expand the compute of OpenAI.
這將擴大 OpenAI 的覆蓋範圍和運算能力。 - And as you know, they are completely compute constrained.
如你們所知，他們完全受限於算力。 - And so AWS, we accelerate EMR.
在 AWS 方面，我們加速 EMR。 - We accelerate SageMaker.
我們加速 SageMaker。 - We accelerate Bedrock.
我們加速 Bedrock。 - NVIDIA’s integrated it really deeply into AWS.
NVIDIA 已經非常深度地整合進 AWS。
Microsoft Azure 與機密運算
- They were our first cloud partner.
他們是我們的第一個雲端合作夥伴。 - Microsoft Azure.
Microsoft Azure。 - NVIDIA’s A100 supercomputer was the first one we built was for NVIDIA.
NVIDIA 的 A100 超級電腦是我們為 NVIDIA 自己打造的第一台。 - The first one we installed was at Azure.
我們安裝的第一台是在 Azure。 - And that led to the big successful partnership with OpenAI.
這促成了與 OpenAI 的重大成功合作夥伴關係。 - But we’ve been working with Azure for quite a long time.
但我們與 Azure 合作已經很長時間了。 - We accelerate Azure cloud now, it’s their AI Foundry, we partner deeply with.
我們現在加速 Azure 雲端，就是他們的 AI Foundry，我們與之深度合作。 - We accelerate Bing search.
我們加速 Bing 搜尋。 - We work with them on Azure regions.
我們與他們在 Azure 區域上合作。 - This is one of the areas that is incredibly important as we continue to expand AI throughout the world.
這是隨著我們持續將 AI 擴展到全世界而變得極其重要的領域之一。 - One of the capabilities that we offer is confidential computing.
我們提供的能力之一是機密運算。 - That in confidential computing, you want to make sure that even the operator cannot see your data, even the operator cannot touch or see your models.
在機密運算中，你要確保即使營運者也看不到你的資料，營運者也無法接觸或看到你的模型。 - Confidential computing and VSGPs use the first ones in the world to do that.
機密運算和 VSGPU 是世界上第一批做到這一點的。 - It’s now able to support confidential computing and protected deployment of these very valuable OpenAI models and and and tropic models throughout clouds.
現在能夠支援機密運算以及在各雲端中受保護地部署這些非常有價值的 OpenAI 模型和 Anthropic 模型。 - And different types of AI, different regions.
以及不同類型的 AI、不同的區域。 - And all because of our account, confidential computing, confidential computing super important.
這一切都因為我們的機密運算能力，機密運算非常重要。
Synopsis、Oracle 與 AI 雲端
- And here’s an example where we have different customers that we work with.
這裡是一個我們與不同客戶合作的案例。 - Synopsis, a great partner of ours, we’re accelerating all of their EDA and CU workflows, and then we landed at Microsoft Azure.
Synopsys，我們的重要合作夥伴，我們正在加速他們所有的 EDA (Electronic Design Automation) 和 CU (Computational Use) 工作流程，然後我們落地到 Microsoft Azure。 - We were Oracle’s first AI customer.
我們是 Oracle 的第一個 AI 客戶。 - Most people would have thought we were their first supplier, we’re their first supplier also, but we were their first AI customer.
大多數人會以為我們是他們的第一個供應商，我們確實也是他們的第一個供應商，但我們也是他們的第一個 AI 客戶。 - I’m quite proud of the fact that I explain AI clouds to Oracle for the first time and we were their first customer.
我很自豪的是，我第一次向 Oracle 解釋了 AI 雲端，而我們是他們的第一個客戶。 - Since then, they’ve really taken off.
從那之後，他們真的起飛了。 - We’ve landed a whole bunch of our partners, their Cohere and Fireworks and of course, very famously, OpenAI.
我們已經讓一大批合作夥伴落地，Cohere 和 Fireworks，當然還有非常知名的 OpenAI。 - A great partnership with Cohere.
與 Cohere 的出色合作夥伴關係。 - Cohere, we’ve, they’re the world’s first AI-native cloud.
Cohere，他們是全球第一個 AI 原生雲端。 - A company that was built with only one singular purpose: to provision, to host GPUs as the era of accelerated computing showed up and the host for AI clouds.
一家只為單一目的而建立的公司：在加速運算時代到來時提供和託管 GPU，並作為 AI 雲端的主機。 - They’ve got some fantastic customers and they’re growing incredibly.
他們有一些很棒的客戶，而且成長驚人。 - One of the platforms that I’m quite excited about is Palantir and Dell.
我非常興奮的平台之一是 Palantir 和 Dell。 - The three of our companies have made it possible to stand up a brand new type of AI platform, the Palantir Ontology platform and AI platform.
我們三家公司合力使建立一種全新類型的 AI 平台成為可能，即 Palantir Ontology 平台和 AI 平台。 - And we could stand up these platforms in any country, in any air-gapped region, completely on-prem, completely on-site, completely in the field.
我們可以在任何國家、任何隔離區域建立這些平台，完全在地端、完全在現場、完全在實地。（Ernest：可惡想要！）
Palantir、Dell 與 AI 平台
- AI could be deployed literally everywhere without our confidential computing capability, without our ability to build the end-to-end system, as well as offer the entire.
如果沒有我們的機密運算能力，沒有我們建構端到端系統的能力，以及提供完整的。 - Accelerated computing and AI stack from data processing whether it’s vectors or structures all the way to AI it would have been possible.
從資料處理到 AI 的完整加速運算和 AI 堆疊，無論是向量還是結構化資料，這一切才成為可能。 - I wanted to show you these examples.
我想向你們展示這些案例。 - This is.
這就是。
NVIDIA 的垂直整合策略
- Our special working relationship with the world’s cloud service providers and many while all of them are here and I get the benefit of seeing them during boot tour and it’s just so incredibly exciting.
我們與全球雲端服務供應商的特殊工作關係，他們都在這裡，我在場外巡視時有幸見到了他們，這真的非常令人興奮。 - I just want to thank all of you for the hard work what Nvidia has done is this.
我只想感謝你們所有人的辛勤工作，NVIDIA 所做的就是這些。 - Are you going to see this theme over and over again.
你們將會一再看到這個主題。 - And videos vertically integrated the world’s first vertically integrated.
NVIDIA 垂直整合，全球第一個垂直整合。 - But horizontally open.
但水平開放。 - And the reason that’s necessary is very simple.
這之所以必要的原因很簡單。 - Accelerated computing is not a chip problem.
加速運算不是晶片問題。 - Accelerated computing is not a systems problem accelerated computing has a missing word we just never say it anymore application acceleration.
加速運算不是系統問題，加速運算有一個被省略的詞，我們只是不再說了：應用加速。 - You can if I could make a computer run everything faster that’s called a CPU.
如果我能讓一台電腦把所有東西都跑得更快，那就叫 CPU。 - But that’s run out of steam.
但那已經後繼無力了。 - The only way for us to accelerate applications going forward and continue to bring.
未來我們加速應用程式並持續帶來的唯一方式。 - Tremendous speed up tremendous cost reduction is through application or domain specific acceleration.
巨大的加速和巨大的成本降低是透過應用或領域專屬的加速來實現的。 - I drop that phrase in the in the front and therefore just became applicant accelerated computing.
我把前面那個詞省略掉了，因此就變成了加速運算。 - And that is the reason why Nvidia has to be library after library domain after domain vertical vertical.
這就是為什麼 NVIDIA 必須一個函式庫接一個函式庫、一個領域接一個領域、一個垂直產業接一個垂直產業。 - We are a vertically integrated computing company.
我們是一家垂直整合的運算公司。 - There is.
沒有。 - No other way.
沒有其他方法。 - We have to understand the applications we have to understand the domain we have to understand fundamentally the algorithms and we have to figure out how to deploy the algorithm.
我們必須理解應用、理解領域、從根本上理解演算法，還必須弄清楚如何部署演算法。 - In whatever scenario it wants to be deployed whether it’s a data center cloud on Prem at the edge or in a robotic system.
無論在什麼場景下部署，無論是資料中心、雲端、地端、邊緣還是機器人系統。 - All of those computing systems are different and finally the systems and chips.
所有這些運算系統都不同，最後才是系統和晶片。 - We are vertically integrated.
我們是垂直整合的。 - What makes it incredibly powerful.
使其具有強大力量的原因。 - And the reason why you saw all the slides.
也是你們看到所有那些投影片的原因。 - That’s because Nvidia is horizontally open.
那是因為 NVIDIA 是水平開放的。 - We’ll work and integrate and videos technology into whatever platform you would like us to integrate into we offer you the software we offer you libraries.
我們會將 NVIDIA 技術整合進你們希望我們整合的任何平台中，我們提供軟體、提供函式庫。 - We integrate with your technology so that we can bring accelerated computing to everybody in the world.
我們與你們的技術整合，這樣我們就能將加速運算帶給全世界的每一個人。
GTC 展示與產業
- Well.
好的。 - This GTC is really a great demonstration of that.
這次 GTC 真的是一個很好的展示。 - You know most of the time.
你們知道大部分時候。 - Most of the time you’ll see me talk about.
大部分時候你們會看到我談論。 - These verticals and I’ll use some examples but in every single case.
這些垂直產業，我會用一些例子，但在每一個案例中。 - Whether it’s automotive by the way financial services the largest percentage of attendees at this GTC is from the financial services industry.
無論是汽車產業，順帶一提，這次 GTC 最大比例的參加者來自金融服務產業。 - I know I’m hoping it’s developers not traders.
我希望是開發者而不是交易員。
生態系與供應鏈
- Guys.
各位。 - Here’s here’s.
這裡。 - Here’s one thing I wanted.
有一件事我想說。 - To say and so.
就是。 - In the audience represents in videos ecosystem.
在場的觀眾代表了 NVIDIA 的生態系。 - Upstream of our supply chain and downstream of our supply chain and we work we think of our supply chain upstream and downstream.
我們供應鏈的上游和下游，我們將供應鏈分為上游和下游。 - And it’s just so exciting.
這真的非常令人興奮。 - That.
那就是。 - Our entire upstream supply chain this last year.
我們整個上游供應鏈在過去一年。 - Irrespective of whether you’re a 50 year old company we have 70 year old companies.
無論你是一家 50 年的公司，我們有 70 年的公司。 - We have a hundred.
我們有 100 年。 - And fifty year old company.
和 150 年的公司。 - Who are now part of the video supply chain and partner with us either upstream or downstream and last year.
它們現在都是 NVIDIA 供應鏈的一部分，在上游或下游與我們合作，去年。 - You had your record year.
你們創下了歷史紀錄。 - Did you not.
不是嗎。 - Congratulations.
恭喜。 - We’re on to something here.
我們正在做一件大事。 - This is the beginning of something very very big.
這是一件非常非常重大的事情的開端。 - And so.
所以。
領域專用函式庫與垂直產業
- If you look at accelerated computing we’ve now set the computing platform.
如果你們看加速運算，我們現在已經建立了運算平台。 - But in order for us.
但為了讓我們。 - To activate those computing platforms we need to have domain specific libraries.
啟動這些運算平台，我們需要領域專用函式庫。 - That solve very important problems in each one of the verticals that we address you see us addressing every single one of this.
在我們所涉及的每一個垂直產業中解決非常重要的問題，你們看到我們在處理每一個。 - Autonomous vehicles.
自動駕駛車輛。 - Are reach our breath.
我們的觸及範圍。 - Our impact incredible we have a track on that financial services I just mentioned algorithmic trading is going from classical.
我們的影響令人難以置信，我們有相關的議程，我剛提到的金融服務，演算法交易正在從傳統的。 - Machine learning.
機器學習。 - With human feature engineering call.
需要人工特徵工程的。 - Quant the quants did that to now.
量化分析師做的那些，到現在。 - Super computers studying massive amounts of data discovering insight and discovering patterns by itself and so this is going through its deep learning and its transformer moment.
超級電腦研究大量資料，自行發現洞見和模式，所以這個領域正在經歷它的深度學習和 Transformer 時刻。 - Healthcare is going is going through their ChatGPT moment some really exciting work that we’re there.
醫療保健正在經歷它們的 ChatGPT 時刻，我們在那裡有一些非常令人興奮的工作。 - We have we have a great keynote track here we have a great keynote track Kimberly pounds in the great keynote track.
我們這裡有一個很棒的主題演講議程，Kimberly Powell 在醫療保健議程中。 - For healthcare.
關於醫療保健。 - We’re talking about AI physics or AI biology.
我們在談論 AI 物理或 AI 生物。 - For drug discovery AI agents for customer service.
用於藥物發現的 AI、用於客戶服務的 AI Agent。 - And support.
和支援。 - Of diagnosis diagnosis and of course physical AI robotic systems.
診斷，以及當然還有物理 AI 機器人系統。 - All these different vectors of AI have different platforms that in video provides industrial we are completely resetting and starting the largest build out of human history and.
所有這些不同的 AI 方向都有 NVIDIA 提供的不同平台，工業方面我們正在完全重建並開始人類歷史上最大的建設。 - Most of the world’s industries.
全球大部分的產業。
產業專屬 AI 部署
- Building AI factories building chip plants building computer plants are represented here today.
建造 AI 工廠、建造晶片廠、建造電腦工廠的企業今天都在這裡。 - Media and entertainment gaming of course real time AI platform.
媒體與娛樂、遊戲，當然還有即時 AI 平台。 - So that we could.
這樣我們就能。 - Translation and broadcast support and live live games and live video.
翻譯和廣播支援，以及即時遊戲和即時影片。 - Enormous amount of it will be augmented with AI we have a we have a platform called Hollis can quantum there. 35 different companies.
大量的內容將透過 AI 增強，我們有一個平台，那裡有量子。35 家不同的公司。 - Here building with us the next generation of quantum GPU hybrid systems.
在這裡與我們一起打造下一代量子 GPU 混合系統。 - Retail and CPG using Nvidia for supply chain using creating a genetic shopping systems.
零售和消費品產業使用 NVIDIA 進行供應鏈管理，建立智慧購物系統。 - AI agents for customer support a lot of work being done here.
用於客戶支援的 AI Agent，這方面有大量工作正在進行。 - 35 trillion dollar industry robotics 50 trillion dollar industry and manufacturing and videos been working in this area for a decade now building 3 computers the fundamental.
35 兆美元的機器人產業、50 兆美元的製造業，NVIDIA 在這個領域已經工作了十年，打造了 3 台基礎電腦。 - Computers necessary to work.
建構機器人系統所需的電腦。 - Build robotic systems we are integrated with working with literally every single company that we know of building robots we have a 110 robots here at the show and then telecommunications.
我們與幾乎每一家我們所知的機器人公司整合合作，我們在展會上有 110 台機器人，然後是電信。 - About as large as the world’s IT industry about 2 trillion dollars.
規模與全球 IT 產業相當，約 2 兆美元。 - We see of course base stations everywhere it’s one of the world’s infrastructures it was the infrastructure of the last generation of computing.
我們當然到處都看到基地台，它是全球基礎設施之一，是上一代運算的基礎設施。 - That infrastructure.
那個基礎設施。 - Is going to get completely reinvented and the reason for that is very simple that base station which is.
將被徹底重新發明，原因很簡單，那個基地台。 - It does one thing which is base station.
它只做一件事，就是充當基地台。
電信基礎設施
- Is going to be an AI infrastructure platform in the future AI will run at the edge.
在未來將成為 AI 基礎設施平台，AI 將在邊緣端執行。 - And so lots of lots of great great discussion there in our platform there is called aerial or a Iran big partnership with Nokia big nut partnership with T-Mobile and many others.
所以有很多很好的討論，我們的平台叫做 Aerial，與 Nokia 有重大合作夥伴關係，與 T-Mobile 和許多其他公司也有重大合作。 - At the core of our business.
在我們業務的核心。 - Everything that I just mentioned.
我剛才提到的一切。
CUDA X 函式庫與演算法
- Computing platforms but very importantly our CUDA X libraries are CUDA X libraries is the algorithm the algorithms and invidia invents we are an algorithm company.
運算平台，但非常重要的是我們的 CUDA X 函式庫，CUDA X 函式庫就是演算法，是 NVIDIA 發明的演算法，我們是一家演算法公司。 - That’s what makes us special that what that’s what makes it possible for me to be able to go into every single one of these industries.
這就是讓我們與眾不同的地方，這讓我能夠進入每一個產業。 - Imagine the future and have the world’s best computer scientists.
想像未來，擁有世界上最優秀的電腦科學家。 - Describe and solve problems.
描述和解決問題。 - Refactor and re express it.
重構並重新表達它。 - And turn it into a library we have so many I think we have.
然後將它變成一個函式庫，我們有很多，我想我們有。 - At this show we’re announcing a hundred.
在這次大會上我們宣布了一百個。 - A hundred libraries.
一百個函式庫。 - So 70 libraries maybe 40 models.
大概 70 個函式庫，或許 40 個模型。 - And that’s just at the show we’re updating these all the time we’re updating them all the time.
這只是在這次大會上，我們一直在更新這些，一直在更新。 - The libraries is the crown jewels of our company it is what makes it possible for that platform the computing platform to be.
函式庫是我們公司的皇冠上的寶石，它使運算平台能夠。 - Activated in service of solving a problem making impact one of the biggest one of the most important libraries that we ever created.
被啟動以服務於解決問題、產生影響，我們有史以來建立的最大、最重要的函式庫之一。 - Who D and N.
cuDNN。 - CUDA deep neural networks it completely revolutionized artificial intelligence cause the big bang of modern AI let me show you a short video about CUDA X. 20 years ago we built CUDA.
CUDA 深度神經網路，它徹底革新了人工智慧，引發了現代 AI 的大爆炸，讓我展示一段關於 CUDA X 的短片。20 年前我們打造了 CUDA。
CUDA 演進與函式庫影響
- A single architecture for accelerated computing.
一個用於加速運算的單一架構。 - Today we’ve reinvented computing a thousand CUDA X libraries help developers make breakthroughs in every field of science and engineering.
今天我們重新發明了運算，一千個 CUDA X 函式庫幫助開發者在每個科學和工程領域取得突破。 - cuOpt for decision optimization.
cuOpt 用於決策最佳化。 - cuLitho for computational lithography.
cuLitho 用於運算微影。 - cuDSS for direct sparse solvers.
cuDSS 用於直接稀疏求解器。 - cuEquivariance for geometry aware neural networks.
cuEquivariance 用於幾何感知神經網路。 - Look around take someplace you’re correctly eliminated.
看看四周，你已經正確地理解了。
模擬與演算法理解
- Dangers.
注意。 - I’m sorry.
抱歉。 - Everything you saw was a simulation.
你們看到的一切都是模擬。 - Some of it was principle solvers.
其中一些是原理求解器。 - Fundamental physics solvers.
基礎物理求解器。 - Some of it was AI surrogate.
其中一些是 AI 代理模型。 - AI physical models, and some of it was physical AI robotics models.
AI 物理模型，還有一些是物理 AI 機器人模型。 - Everything was simulated.
一切都是模擬的。 - Nothing was animated.
沒有任何東西是動畫。 - Nothing was articulated.
沒有任何東西是手動操作的。 - Everything was completely simulated.
一切都是完全模擬的。 - That is what fundamentally NVIDIA does.
這就是 NVIDIA 從根本上所做的事情。
垂直整合與開放性
- It is through the connection of understanding of the algorithms with our computing platforms that we’re able to open up to unlock these opportunities.
正是透過將對演算法的理解與我們的運算平台相連結，我們才能夠開啟並解鎖這些機會。 - NVIDIA is a vertically integrated computing company with open horizontal integration with the world.
NVIDIA 是一家垂直整合的運算公司，同時與全世界進行開放的水平整合。
2️⃣ 推論轉折點
AI 生態系：成熟企業與原生公司
- So that’s CUDA X.
以上就是 CUDA X。 - Well, just now you saw a whole bunch of companies.
剛才你們看到了一大堆公司。 - You saw Walmart and, you know, there’s L’Oreal and incredible companies, established companies, JP Morgan and Roche.
你們看到了 Walmart，還有 L'Oreal 以及許多優秀的成熟企業，像是 JP Morgan 和 Roche。 - These are companies that define society today.
這些都是定義當今社會的公司。 - Toyota.
Toyota。 - These are some of the largest companies in the world.
這些是全世界規模最大的公司。 - It is also true that there’s a whole bunch of companies you’ve never heard of.
同樣也有一大堆你從未聽過的公司。 - These are companies, we call them AI natives, a whole bunch of small companies.
這些公司，我們稱之為 AI 原生公司，一大堆小型公司。 - The list is gigantic.
名單非常龐大。 - This is just a little tiny bit of it, and I couldn’t decide whether to show you more or show you less, and so I made it so that you couldn’t see any.
這只是其中極小的一部分，我沒辦法決定要多放還是少放，所以我就讓大家什麼都看不清楚。 - And nobody’s feelings are hurt.
這樣就沒有人會受傷了。 - However, inside this list are a bunch of brand new companies.
不過，在這份名單裡有一堆全新的公司。 - They’re companies like, for example, you might have heard a couple of them, OpenAI, Anthropic, but there’s a whole bunch of others.
例如，你們可能聽過其中幾家，OpenAI、Anthropic，但還有一大堆其他的。 - There’s a whole bunch of others, and they serve different verticals.
還有很多很多，而且它們服務不同的垂直產業。
投資熱潮與 AI 原生公司
- Something happened in the last two years, particularly this last year.
過去兩年發生了一些事，尤其是這最近一年。 - We’ve been working with the AI natives for a long time, and this last year it just skyrocketed.
我們與 AI 原生公司合作已經很長一段時間了，而去年一切突然暴增。 - I’ll explain to you why it happened.
我會跟你們解釋為什麼會這樣。 - This is a big deal.
這是一件大事。 - The industry has skyrocketed, $150 billion of investment into venture investment, into startups, the largest in human history.
這個產業已經飛速成長，1,500 億美元的創投資金投入新創公司，是人類歷史上最大的規模。 - This is also the first time that the scale of the investments went from millions of dollars, tens of millions of dollars, to hundreds of millions of dollars and billions of dollars.
這也是投資規模第一次從數百萬美元、數千萬美元，跳升到數億美元和數十億美元。 - And the reason for that is this is the first time in history that every single one of these companies needs a company.
原因在於，這是歷史上第一次每一家這樣的公司都需要運算能力。 - They need compute and lots and lots of it.
它們需要運算，而且需要非常、非常多。 - They need tokens, lots and lots of it.
它們需要 token，而且需要非常、非常多。 - They’re either going to create and build and create tokens and generate tokens, or they’re going to integrate, add value to tokens that are available, created by Anthropic and OpenAI and others.
它們要不就是自己建立和生成 token，要不就是整合由 Anthropic、OpenAI 等公司所產出的 token 並加以增值。 - And so this industry is different in so many different ways, but the one thing that is very clear.
所以這個產業在很多方面都不同，但有一件事非常清楚。 - The impact that they’re making, the incredible value that they’re delivering already is quite tangible.
它們正在產生的影響力，它們已經在提供的驚人價值，是相當具體可見的。 - AI natives.
AI 原生公司。
重新發明運算與生成式 AI
- All because we reinvented computing.
這一切都是因為我們重新發明了運算。 - Just like during the PC revolution, a whole bunch of new companies were created.
就像在 PC 革命時期，一大堆新公司被建立出來。 - Just as during the Internet revolution, a whole bunch of companies were created, and in mobile cloud, a whole bunch of companies were created.
就像在網際網路革命時期，一大堆公司被建立出來，在行動雲端時代，也是一大堆公司被建立出來。 - Each one of them had their own standards, and we’re talking about one of the major standards that just happened, incredibly important.
每一次都有自己的標準，而我們正在談論的是剛剛才發生的一個重要標準，極為重要。 - And this generation, we also have our own large number of very, very special companies.
在這個世代，我們也有一大批非常、非常特別的公司。 - We reinvented computing.
我們重新發明了運算。 - It stands to reason there’s going to be a whole new crop of really important companies, consequential companies for the future of the world.
理所當然地，會有一批全新的、真正重要的公司出現，對世界未來具有深遠影響的公司。 - The Googles, the Amazons, the Metas, consequential companies that have come as a result of the last computing platform shift.
Google、Amazon、Meta，這些都是上一次運算平台轉移所產生的具有深遠影響力的公司。 - We are now at the beginning of a new platform shift.
我們現在正處於一次新的平台轉移的開端。 - But what happened in the last couple of years?
但過去幾年發生了什麼？ - Well, we’ve been watching, as you know, we’ve been working on deep learning and working on AI.
如你們所知，我們一直在關注，我們一直在深度學習和 AI 上努力。 - The big bang of modern AI, we were right there at the spot, and we’ve been advancing this field for quite some time.
現代 AI 的大爆發，我們就在那個現場，而且我們已經推動這個領域相當長一段時間了。 - But why the last two years?
但為什麼是這最近兩年？ - What happened in the last two years?
這兩年發生了什麼？ - Well, three things.
有三件事。 - ChatGPT, of course, started the generative AI era.
ChatGPT 當然開啟了生成式 AI 的時代。 - It’s able to not just understand, perceive and understand.
它不只能理解，能感知和理解。 - It’s able to also translate and generate, generation of unique content, I showed you the fusion of generative AI with computer graphics, and it brought computer graphics to life.
它還能翻譯和生成，生成獨特的內容，我剛才展示了生成式 AI 與電腦圖學的融合，它讓電腦圖學活了過來。 - You guys, everybody in the world should be using ChatGPT.
各位，全世界每個人都應該使用 ChatGPT。 - I know I use it every single morning.
我知道我每天早上都在用。 - I used it plenty this morning.
今天早上我就用了很多次。 - And so ChatGPT was the generative AI era.
所以 ChatGPT 就是生成式 AI 時代。 - The second, by the way, generative computing versus the way we used to do computing, it’s not, generative AI is a capability of software.
第二，順帶一提，生成式運算相對於我們過去的運算方式，生成式 AI 是軟體的一種能力。 - But it has profoundly changed, it has completely changed how computing is done.
但它已經深刻地改變了，它已經完全改變了運算的方式。 - Computing used to be retrieval based, now it’s generative.
運算過去是基於檢索的，現在是生成式的。 - Keep that thought in mind when I talk about certain things.
當我接下來談到某些事情時，請記住這個概念。 - And you’ll realize why it is that everything that we do is going to change how computers are architected, how computers are provided.
你就會理解為什麼我們所做的一切將改變電腦的架構方式、電腦的提供方式。 - How computers are going to be built out, and what is the meaning of computing altogether.
電腦將如何被建構，以及運算的整體意義是什麼。
生成式、推理式與自主式 AI
- Generative AI, 2023, end of 22, 2023.
生成式 AI，2023 年，2022 年底到 2023 年。 - Reasoning AI, o1, which, and then took off with o3.
推理式 AI，o1，然後在 o3 之後真正起飛。 - Reasoning allowed it to reflect, allows it to think to itself, allowed it to plan, break down problems and decompose a problem it couldn’t understand into steps or parts that it could understand.
推理讓它能夠反思，讓它能夠自我思考，讓它能夠規劃，分解問題，把一個它無法理解的問題拆解成它能理解的步驟或部分。 - It could ground itself on research. o1 made generative AI trustworthy and grounded on truth.
它能夠以研究為基礎進行驗證。o1 讓生成式 AI 變得值得信賴，並以事實為根基。 - That caused ChatGPT to simply took off, and that was a very, very big moment.
這讓 ChatGPT 直接起飛，那是一個非常、非常重大的時刻。 - The amount of input tokens that was necessary in order to produce and the amount of output tokens it generated in order to reason, the model was a little bit larger.
為了產出所需的輸入 token 數量，以及為了推理而生成的輸出 token 數量，模型變得稍微大了一些。 - You know, of course, you could have much larger models.
當然，你可以擁有更大的模型。 - o1 was a little bit larger, not much larger, but its input token usage for context and its output token for context and for thinking increased the amount of computation tremendously.
o1 稍微大了一點，不是大很多，但它在上下文的輸入 token 使用量和用於上下文及思考的輸出 token 大幅增加了運算量。 - Then came Claude Code, the first agentic model.
然後是 Claude Code，第一個自主式模型。 - It was able to read files, code, compile it, test it, evaluate it, go back and iterate on it.
它能夠讀取檔案、寫程式、編譯、測試、評估，然後回頭迭代改進。 - Claude Code has revolutionized software engineering, as all of you know. 100% of NVIDIA is using a combination of, or oftentimes all three of them, Claude Code, Codex, and Cursor, all over NVIDIA.
如大家所知，Claude Code 已經徹底改變了軟體工程。NVIDIA 百分之百都在使用其中的組合，或者經常三個都用，Claude Code、Codex 和 Cursor，遍及整個 NVIDIA。 - There’s not one software engineer today who is not assisted by one or many AI agents helping them code.
今天沒有一個軟體工程師不是由一個或多個 AI 代理來協助寫程式的。 - Claude Code completely revolutionizes the new inflection.
Claude Code 完全帶來了新的轉折點革命。 - And for the first time, you don’t ask an AI what, where, when, how.
而且第一次，你不再問 AI 什麼、哪裡、何時、如何。 - You ask it create, do, build, you ask it to use tools, take your context, read files.
你叫它建立、執行、打造，你叫它使用工具、接收你的上下文、讀取檔案。 - It’s able to agentically break down a problem, reason about it, reflect on it.
它能夠自主地分解問題、對其進行推理、對其進行反思。 - It’s able to solve problems and actually perform tasks.
它能夠解決問題並實際執行任務。 - An AI that was able to perceive became an AI that could generate.
一個能夠感知的 AI 變成了一個能夠生成的 AI。 - An AI that could generate became an AI that could reason.
一個能夠生成的 AI 變成了一個能夠推理的 AI。 - An AI that could reason now became an AI that can actually do work.
一個能夠推理的 AI 現在變成了一個能夠真正完成工作的 AI。 - Very productive work.
非常有生產力的工作。
推論轉折點
- The amount of computation in the last two years, we know that everybody in this room knows, the computing demand for NVIDIA GPUs is off the charts.
過去兩年的運算量，我們知道在座所有人都知道，對 NVIDIA GPU 的運算需求已經超乎想像。 - Spot pricing is skyrocketing.
現貨價格正在飆升。 - You couldn’t find a GPU if you tried, and yet in the meantime, we’re shipping GPUs out.
就算你想找也找不到 GPU，但與此同時，我們正在不斷出貨 GPU。 - Incredible amounts of it.
數量驚人。 - And demand just keeps on going up.
而需求就是不斷往上攀升。 - There’s a reason for that.
這是有原因的。 - This fundamental inflection.
這個根本性的轉折點。 - Finally, AI is able to do productive work, and therefore, the inflection point of inference has arrived.
AI 終於能夠做有生產力的工作了，因此，推論的轉折點已經到來。 - AI now has to think.
AI 現在必須思考。 - In order to think, it has to inference.
為了思考，它必須進行推論。 - AI now has to do.
AI 現在必須行動。 - In order to do, it has to inference.
為了行動，它必須進行推論。 - AI has to read.
AI 必須閱讀。 - In order to do so, it has to inference.
為了做到這一點，它必須進行推論。 - It has to reason.
它必須推理。 - It has to inference.
它必須進行推論。 - Every part of AI, every time it has to think, it has to reason, it has to do, it has to generate tokens, it has to inference.
AI 的每一個部分，每一次它必須思考、推理、行動、生成 token，它都必須進行推論。 - It’s way past training now.
現在已經遠遠超越訓練階段了。 - It’s in the field of inference.
它已經進入推論的領域。 - So the inference inflection has arrived.
所以推論的轉折點已經到來。
大規模運算需求與營收預測
- At the time when the amount of tokens, the amount of compute necessary, increased by roughly 10,000 times.
在 token 數量、所需的運算量增加了大約 10,000 倍的時候。 - Now, when I combine these two, the fact that since the last two years, the computing demand of the work has gone up by 10,000 times.
現在，當我把這兩者結合起來，過去兩年工作的運算需求增加了 10,000 倍。 - And the amount of usage, the amount of usage has probably gone up by 100 times.
而使用量，使用量大概增加了 100 倍。 - People have heard me say, I believe that computing demand has increased by 1 million times in the last two years.
大家都聽過我說，我相信過去兩年運算需求增加了 100 萬倍。 - It is the feeling that we all have.
這是我們所有人的感受。 - It is the feeling every startup has.
這是每一家新創公司的感受。 - It’s the feeling that OpenAI has.
這是 OpenAI 的感受。 - It’s the feeling that Anthropic has.
這是 Anthropic 的感受。 - If they could just get more capacity, they could generate more tokens.
如果他們能獲得更多容量，就能生成更多 token。 - Their revenues would go up.
他們的營收就會上升。 - More people could use it.
更多人就能使用。 - The more advanced, the smarter the AI could become.
AI 就能變得越先進、越聰明。 - We are now at that positive flywheel system.
我們現在正處於那個正向飛輪系統中。 - We have reached that moment.
我們已經到達了那個時刻。 - The inflection, the inference inflection has arrived.
轉折點，推論的轉折點已經到來。 - Last year, at this time, I said, that where I stood at that moment in time, we saw about $500 billion.
去年這個時候，我說，在我當時所站的位置，我們看到了大約 5,000 億美元。 - We saw $500 billion of very high confidence demand and purchase orders for Blackwell and Rubin through 2026.
我們看到了 5,000 億美元非常高信心度的需求和 Blackwell 及 Rubin 直到 2026 年的採購訂單。 - I said that last year.
我去年這樣說的。 - Now, I don’t know if you guys feel the same way, but $500 billion is an enormous amount of revenue.
我不知道大家是否有同樣的感覺，但 5,000 億美元是一筆巨大的營收。 - Not one impressed.
沒有一個人覺得驚訝。 - I know why you’re not impressed, because all of you had record years.
我知道為什麼你們不覺得驚訝，因為你們每個人都創下了歷史紀錄。 - Well, I’m here to tell you that right now where I stand, a few short months after GTCDC, one year after last GTC, right here where I stand, I see, through 2027, at least $1 trillion.
我在這裡要告訴你們，就在我現在所站的位置，距離 GTCDC 僅僅幾個月後，距離上次 GTC 一年後，就在我站的這裡，我看到到 2027 年，至少 1 兆美元。
推論聚焦與通用 AI 平台
- Now, does it make any sense?
這合理嗎？ - And that’s what I’m going to spend the rest of the time talking about.
這就是我接下來要花時間談論的內容。 - In fact, we are going to be short.
事實上，我們的供應還會不足。 - I am certain computing demand will be much higher than that.
我確信運算需求會遠遠高於那個數字。 - And there’s a reason for that.
這是有原因的。 - So the first thing is, we did a lot of work in the last year.
第一件事是，我們在過去一年做了大量的工作。 - Of course, as you know, 2025 was NVIDIA’s year of inference.
如你們所知，2025 年是 NVIDIA 的推論之年。 - We wanted to make sure that not only were we good at training and post-training, that we were incredibly good at every single phase of AI.
我們想要確保我們不僅在訓練和後訓練方面表現優異，而是在 AI 的每一個階段都表現得非常出色。 - So that the investments that were made, investments made in our infrastructure, could scale out for as long as they would like to use it.
這樣所做的投資，對我們基礎設施的投資，就能在他們想要使用的期間持續擴展。 - And the useful life of NVIDIA’s infrastructure would be long, and therefore the cost would be incredibly low.
而且 NVIDIA 基礎設施的有效使用壽命會很長，因此成本會非常低。 - The longer you could use it, the lower the cost.
你能使用得越久，成本就越低。（Ernest: NVIDIA 將 Amazon 飛輪拿來用了！） - There’s no question in my mind, NVIDIA systems are the lowest cost infrastructure you could get for AI infrastructure in the world.
我毫不懷疑，NVIDIA 系統是全世界你能取得的最低成本 AI 基礎設施。 - And so the first part was, last year was all about AI for inference.
所以第一部分是，去年全都是關於 AI 推論。 - And it drove this inflection point.
這推動了這個轉折點。 - Simultaneously, we were very pleased last year that Anthropic has come to NVIDIA.
同時，我們非常高興去年 Anthropic 加入了 NVIDIA 的陣營。 - That MSL, MetaSL, has chosen NVIDIA.
MSL、MetaSL 選擇了 NVIDIA。 - And meanwhile, meanwhile, and as a collection, as a group, this represents one third of the world’s AI compute.
而且與此同時，作為一個整體，一個群體，這代表了全球三分之一的 AI 運算。 - Open source models.
開源模型。 - Open source models have reached near the frontier, and it is literally everywhere.
開源模型已經接近前沿水準，而且它真的無處不在。 - And NVIDIA, as you know, today, we’re the only platform in the world today that runs every single domain of AI across every single one of these AI models.
如你們所知，NVIDIA 今天是全世界唯一一個能跨所有 AI 模型執行每一個 AI 領域的平台。 - In language, in biology, in computer graphics, computer vision, in speech, proteins and chemicals, robotics and otherwise, edge or cloud, any language.
在語言、生物學、電腦圖學、電腦視覺、語音、蛋白質和化學、機器人等領域，邊緣或雲端，任何語言。 - NVIDIA’s architecture is fungible for all of that, and we’re incredible for all of that.
NVIDIA 的架構對所有這些都是通用的，而且我們在所有這些方面都表現出色。 - That allows us to be the lowest cost, the highest confidence platform.
這使我們成為最低成本、最高信心度的平台。 - Because when you’re building these systems, as I mentioned, a trillion dollars is an enormous amount of infrastructure.
因為當你在建造這些系統時，如我所提到的，1 兆美元是非常龐大的基礎設施投資。 - You have to have complete confidence, that the trillion dollars you’re putting down will be utilized, would be performant, would be incredibly cost effective.
你必須完全有信心，你投入的那 1 兆美元會被充分利用、會有高效能、會非常具有成本效益。 - And have useful life for as long as you could see.
而且在可見的未來都能持續發揮效用。 - That infrastructure investment you could make on NVIDIA, you could make with complete confidence.
你在 NVIDIA 上的基礎設施投資，你可以完全有信心地進行。 - We have now proven that.
我們現在已經證明了這一點。 - It is the only infrastructure in the world that you could go anywhere in the world and build with complete confidence.
這是全世界唯一一個你可以在世界任何地方以完全的信心來建造的基礎設施。 - You want to put it in any of the companies, you want to put it in any of the clouds, we’re delighted by that.
你想把它放在任何公司、任何雲端，我們都很高興。 - You want to put it on-prem, we’re happy about that.
你想要放在地端，我們也很高興。 - You want to put it in any country anywhere, we’re delighted to support you.
你想要放在任何國家的任何地方，我們都很樂意支援你。 - We are now a computing platform that runs all of AI.
我們現在是一個能夠執行所有 AI 的運算平台。
業務區隔與 AI 韌性
- Now, our business already starting to show that. 60% of our business is hyperscalers, the top five hyperscalers.
我們的業務已經開始展現出這一點。我們業務的 60% 來自超大規模業者，前五大超大規模業者。 - However, even within that top five hyperscalers, some of it is internal AI consumption.
然而，即使在這前五大超大規模業者中，部分也是內部 AI 消耗。 - The internal AI consumption really important work, like Rexis is moving from recommender systems of tables and collaborative filtering and content filtering.
內部 AI 消耗是非常重要的工作，例如推薦系統正在從表格式的協同過濾和內容過濾轉型。 - It’s moving towards deep learning and large language models.
它正在朝向深度學習和大型語言模型發展。 - Search, moving to deep learning, large language models.
搜尋，正在轉向深度學習、大型語言模型。 - Almost all of these different hyperscale workloads are now moving, shifting towards a workload that NVIDIA GPUs are incredibly good at.
幾乎所有這些不同的超大規模工作負載現在都在轉移，朝向 NVIDIA GPU 極為擅長的工作負載。 - But on top of that, because we work with every AI lab, because we accelerate every AI model, and because we have a large ecosystem of AI natives that we work with, that we can bring to the clouds.
此外，因為我們與每一個 AI 實驗室合作，因為我們加速每一個 AI 模型，因為我們有龐大的 AI 原生公司生態系與我們合作，我們可以帶到雲端上。 - That investment, no matter how large, no matter how quick, that compute will be consumed.
那些投資，無論多大、無論多快，那些運算都會被消耗掉。 - And that represents 60% of our business.
這代表了我們 60% 的業務。 - The other 40% is just everywhere.
另外 40% 就是無所不在。 - Regional clouds, sovereign clouds, enterprise, industrial, robotics, edge, big systems, super computing systems, small servers, enterprise servers, the number of systems, incredible.
區域雲端、主權雲端、企業、工業、機器人、邊緣、大型系統、超級運算系統、小型伺服器、企業伺服器，系統數量之多令人難以置信。 - The diversity of AI is also its resilience.
AI 的多樣性也是它的韌性。 - The span of reach of AI is its resilience.
AI 的觸及範圍就是它的韌性。 - There is no question this is not a one app technology.
毫無疑問，這不是單一應用的技術。 - This is now fundamental.
這現在已經是基礎性的。 - This is absolutely a new computing platform shift.
這絕對是一次新的運算平台轉移。
推論之年：硬體創新
- Well, our job is to continue to advance the technology.
我們的工作就是持續推進技術。 - And one of the most important things that I mentioned last year was last year was our year of inference.
我去年提到的最重要的事情之一就是，去年是我們的推論之年。 - We dedicated everything.
我們投入了一切。 - We took a giant chance and reinvented while Hopper was at its prime, and it was just cooking.
我們冒了一個巨大的風險，在 Hopper 正值巔峰、火力全開的時候進行了重新發明。 - We decided that the Hopper architecture, the NVLink by 8, had to be taken to the next level.
我們決定 Hopper 架構，NVLink 8 路互連，必須提升到下一個層級。 - We completely re-architected the system, disaggregated the computing system altogether, and created NVLink 72.
我們完全重新架構了系統，將運算系統徹底解耦，並建立了 NVLink 72。 - The way that it’s built, the way it’s manufactured, the way it’s programmed, completely changed.
它的建造方式、製造方式、程式設計方式，全部都改變了。 - Grace Blackwell NVLink 72 was a giant bet.
Grace Blackwell NVLink 72 是一個巨大的賭注。 - And it wasn’t easy for anybody.
對任何人來說都不容易。 - And many of my partners here in the room, I want to thank all of you for the hard work that you guys did.
在座的許多合作夥伴，我要感謝你們所有人付出的辛勤努力。 - Thank you.
謝謝。 - Thank you.
謝謝。 - NVLink 72.
NVLink 72。 - NVFP4.
NVFP4。 - Not just FP4 precision.
不僅僅是 FP4 精度。 - FP4 is a whole different type of tensor core and computational unit.
FP4 是一種完全不同類型的 tensor core 和運算單元。 - We’ve demonstrated now that we can inference NVFP4 without loss of precision, but gigantic boost in performance and energy efficiency.
我們現在已經證明，我們可以用 NVFP4 進行推論而不損失精度，但效能和能源效率卻獲得巨大提升。 - We’ve also been able to use NVFP4 for training.
我們也已經能夠將 NVFP4 用於訓練。 - So, NVLink 72.
所以，NVLink 72。 - NVFP4.
NVFP4。 - The invention of Dynamo, Tensor RT LLM, a whole bunch of new algorithms.
Dynamo 的發明、TensorRT-LLM，還有一大堆新的演算法。 - We even built a supercomputer to help us optimize kernels and help us optimize our complete stack.
我們甚至建造了一台超級電腦來幫助我們最佳化核心程式和整個軟體堆疊。 - We call it DGX Cloud.
我們稱之為 DGX Cloud。 - We invested billions of dollars of supercomputing capability to help us create the kernels, the software that made inference possible.
我們投入了數十億美元的超級運算能力來幫助我們建立那些核心程式、那些讓推論成為可能的軟體。 - Well, the results all came together, and people used to tell me, but Jensen, inference is so easy.
成果都匯集在一起了，以前人們告訴我，但是 Jensen，推論很簡單啊。 - Inference is the ultimate hard.
推論是終極的難題。 - Inference is ultimate hard.
推論就是終極的難題。 - It is also ultimate important because it drives your revenues.
它也是終極重要的，因為它驅動你的營收。 - And so this is the outcome.
這就是成果。 - This is from semi-analysis.
這是來自 SemiAnalysis 的資料。 - This is the largest, most comprehensive sweep of AI inference that has ever been done.
這是有史以來最大規模、最全面的 AI 推論評測。 - And what you see here on the left, on this side, on this side, is tokens per watt.
你在這邊左側看到的，是每瓦 token 數。 - Tokens per watt is important because every data center, every single factory, by definition, is power constrained.
每瓦 token 數很重要，因為每一個資料中心、每一座工廠，從定義上來說，都受到電力限制。 - A one gigawatt factory will never become two.
一座 1 GW 的工廠永遠不會變成 2 GW。 - It’s physically constrained.
它受到物理限制。 - The laws of atoms, the laws of physicality.
原子的法則，物理的法則。 - And so that one gigawatt of data center, you want to drive the maximum number of tokens, which is the production, the product of that factory.
所以那一座 1 GW 的資料中心，你要驅動最多的 token 數量，那就是那座工廠的產出、產品。 - So, you want to be on top of that curve as high as you want.
所以，你希望在那條曲線上越高越好。
效能、成本與 Token 工廠
- This, the x-axis, is the interactivity, the speed of inference, the speed of each inference.
X 軸是互動性，推論的速度，每次推論的速度。 - The faster you can inference, the faster you could, of course, respond.
你推論得越快，當然就能越快回應。 - But very importantly, the faster you can inference, the larger the models, the more context you could process, the more tokens you can think through.
但非常重要的是，你推論得越快，就能使用越大的模型、處理越多的上下文、思考越多的 token。 - This axis is the same as smartness of the AI.
這個軸等同於 AI 的聰明程度。 - And so, this is the throughput of the AI.
所以，這是 AI 的吞吐量。 - This is the smartness of the AI.
這是 AI 的聰明程度。 - Notice, the smarter the AI, the lower your throughput.
請注意，AI 越聰明，你的吞吐量就越低。 - Makes sense.
合理。 - You’re thinking longer.
你在思考更久。 - Okay?
對吧？ - And so, this axis is the speed, and I’m going to come back to this.
所以，這個軸是速度，我稍後會回來談這個。 - This is important.
這很重要。 - This is where I torture all of you.
這就是我折磨各位的地方。 - But it’s too important.
但它太重要了。 - Every CEO in the world, you watch, every CEO in the world will study their business from now on in the way I’m about to describe.
全世界的每一位 CEO，你們看著，從現在開始每一位 CEO 都會用我即將描述的方式來研究他們的業務。 - Because this is your token factory.
因為這就是你的 token 工廠。 - This is your AI factory.
這就是你的 AI 工廠。 - This is your revenues.
這就是你的營收。 - There’s no question about that going forward.
往後毫無疑問就是這樣。 - And so, this is the throughput.
所以，這是吞吐量。 - This is the intelligence.
這是智慧程度。 - Better per watt for a given power of data center, the more throughput, the more tokens you could produce.
在既定電力的資料中心裡，每瓦效能越好，吞吐量越高，你能產出的 token 就越多。 - On this side is cost.
這邊是成本。 - Notice, NVIDIA is the highest performance in the world.
請注意，NVIDIA 是全世界最高效能的。 - Nobody would be surprised by that.
沒有人會對此感到驚訝。 - They would be surprised by the fact that in one generation, whereas Moore’s Law would have given us, through transistors, 50%, two times.
他們會驚訝的是，在一個世代中，摩爾定律透過電晶體能給我們 50%、兩倍的提升。 - Moore’s Law would probably give us one and a half times more performance.
摩爾定律大概能給我們 1.5 倍的效能提升。 - You would have expected from Hopper H200, one and a half times higher.
你從 Hopper H200 預期的提升大概是 1.5 倍。 - Nobody would have expected 35 times higher.
沒有人會預期到 35 倍的提升。 - I said last year, at this time, that NVIDIA’s Grace Blackwell, NVLink 72, was 35 times perf per watt.
我去年這個時候說，NVIDIA 的 Grace Blackwell NVLink 72 的每瓦效能是 35 倍。 - Nobody believed me.
沒有人相信我。 - And then, SemiAnalysis came out, and Dylan Patel had a quote.
然後 SemiAnalysis 公布了結果，Dylan Patel 說了一句話。 - He accused me of sandbagging.
他指控我低報數字。 - He says, Jensen sandbagged.
他說，Jensen 低報了。 - It’s actually 50 times.
實際上是 50 倍。 - And he’s not wrong.
他沒有說錯。 - He’s not wrong.
他沒有說錯。 - And so, our cost per token is the lowest in the world.
所以，我們的每 token 成本是全世界最低的。 - You can’t beat it.
你無法超越。 - I’ve said before, if you have the wrong architecture, even if it’s free, it’s not cheap enough.
我之前說過，如果你的架構不對，即使是免費的，也不夠便宜。(Ernest: 噴錢出去之前，請架構優先啊，各位朋友，歡迎來約。) - And the reason for that is because no matter what happens, you still have to build a gigawatt data center.
原因在於，無論如何你仍然必須建造一座 1 GW 的資料中心。 - You still have to build a gigawatt factory.
你仍然必須建造一座 1 GW 的工廠。 - And that gigawatt factory, for $40,000, it’s not cheap.
而那座 1 GW 的工廠，花 4 萬美元，可不便宜。 - It’s not cheap.
不便宜。 - But that gigawatt factory, for 15 years, amortized across, that gigawatt factory is about $40 billion.
但那座 1 GW 的工廠，攤提 15 年，那座工廠大約是 400 億美元。 - Even when you put nothing on it, it’s $40 billion in.
即使你什麼都不放上去，就已經投入了 400 億美元。 - You better make for darn sure you put the best computer system on that thing so that you could have the best token cost.
你最好確定你在上面放了最好的電腦系統，這樣你才能有最好的 token 成本。 - NVIDIA’s token cost is world-class.
NVIDIA 的 token 成本是世界級的。 - Basically, untouchable at the moment.
基本上，目前無人能及。
NVIDIA 的 Token 協同設計策略
- And the reason that’s true is because of extreme co-design.
這之所以成立，是因為極致的協同設計。 - NVIDIA’s token co-design is an observation where the effectiveness, the performance, and the token cost production capability for their factories is everything to them.
NVIDIA 的 token 協同設計是一種觀察，其中工廠的效率、效能和 token 成本生產能力對他們來說就是一切。 - And this is what happened.
而這就是發生的事情。 - This is, we updated their software, same system, and notice their token speeds.
這是，我們更新了他們的軟體，同一套系統，請注意他們的 token 速度。 - Incredible.
令人難以置信。 - The difference before, before NVIDIA updated everything and all of our algorithms and software and all the technology that we bring to bear.
在 NVIDIA 更新一切之前，所有我們的演算法、軟體和我們帶來的所有技術之前的差異。 - About 700 tokens per second average went to nearly 5,000, seven times higher.
平均約每秒 700 個 token 提升到近 5,000 個，提高了 7 倍。 - And so this is the incredible power of extreme co-design.
這就是極致協同設計的驚人威力。 - I mentioned earlier the importance of factories.
我之前提到了工廠的重要性。 - This is the importance of factory.
這就是工廠的重要性。 - Your data center, it used to be a data center for files.
你的資料中心，過去是一個存放檔案的資料中心。
資料中心化身 Token 工廠
- It’s now a factory to generate tokens.
它現在是一座生成 token 的工廠。 - Your factory is limited no matter what.
你的工廠無論如何都是有限的。 - Everybody is looking for land, power, and shell.
每個人都在尋找土地、電力和廠房外殼。 - Once you build it, you are power limited.
一旦建好了，你就受到電力限制。 - Within that power limited infrastructure, you better make for darn sure that your inference, because you know inference is your workload and tokens is your new commodity.
在那個電力受限的基礎設施中，你最好確保你的推論效能，因為你知道推論是你的工作負載，而 token 是你的新商品。 - That compute is your revenues.
那些運算就是你的營收。 - That you want to make sure that the architecture is as optimized as you can.
你要確保架構盡可能最佳化。 - In the future.
在未來。 - Every single CSP.
每一家 CSP。 - Every single computer company, every single cloud company, every single AI company, every single.
每一家電腦公司、每一家雲端公司、每一家 AI 公司、每一家。 - Company period.
所有公司。 - Are going to be thinking about their token factory effectiveness.
都會思考他們的 token 工廠效率。 - This is your factory in the future, and the reason why I know that is because everybody in this room is powered by intelligence.
這就是你未來的工廠，而我之所以知道這一點，是因為在座的每個人都是由智慧所驅動的。 - And in the future that intelligence will be augmented by tokens.
在未來，那個智慧將會被 token 所增強。 - So let me show you how we got here.
讓我展示我們是如何走到這裡的。 - On April 6th, 2016, a decade ago, we introduced DGX one, the world’s first computer designed for deep learning.
2016 年 4 月 6 日，十年前，我們推出了 DGX-1，全世界第一台專為深度學習設計的電腦。 - Eight Pascal GPUs connected with the first generation and V-link 170 teraflops in one computer, the world’s first computer designed for AI researchers.
8 顆 Pascal GPU 透過第一代 NVLink 連接，一台電腦 170 teraflops，全世界第一台專為 AI 研究人員設計的電腦。 - With Volta, we introduced NVLink switch. 16 GPUs connected with full all to all bandwidth operating as one giant GPU.
在 Volta 時代，我們推出了 NVLink Switch。16 顆 GPU 以全 all-to-all 頻寬連接，作為一顆巨大的 GPU 運作。 - A giant step forward, but model sizes continued to grow.
一個巨大的進步，但模型規模持續增長。 - The data center needed to become a single unit of computing.
資料中心需要成為一個單一的運算單元。 - So Mellanox joined Nvidia.
所以 Mellanox 加入了 NVIDIA。 - In 2020, DGX A100 SuperPod became the first GPU supercomputer combining scale up and scale out architecture.
2020 年，DGX A100 SuperPod 成為第一台結合 scale-up 和 scale-out 架構的 GPU 超級電腦。 - NVLink three for scale up, connect X six and quantum and finna band for scale out.
NVLink 3 用於 scale-up，ConnectX-6 和 Quantum InfiniBand 用於 scale-out。 - Then hopper, the first GPU with the FPA transformer engine that launched the generative AI era.
然後是 Hopper，第一顆搭載 FP8 Transformer Engine 的 GPU，開啟了生成式 AI 的時代。 - NVLink for ConnectX-7, BlueField-3 DPU, second generation Quantum InfiniBand be a revolutionized computing.
NVLink for ConnectX-7、BlueField-3 DPU、第二代 Quantum InfiniBand 徹底革新了運算。 - Blackwell redefined AI supercomputing system architecture with envy link 72 72 GPUs connected by envy link spine 130 terabytes per second of all to all bandwidth.
Blackwell 以 NVLink 72 重新定義了 AI 超級運算系統架構，72 顆 GPU 透過 NVLink Spine 連接，130 TB/s 的 all-to-all 頻寬。 - Compute trace integrate Blackwell GPU grace CPUs connect X eight and bluefield three. scale out runs over spectrum for Ethernet.
運算緊密整合 Blackwell GPU、Grace CPU、ConnectX-8 和 BlueField-3。scale-out 透過 Spectrum-X 乙太網路執行。 - With three scaling laws and full.
隨著三個擴展法則和完整的。 - Pre training post training and inference and now a genetic systems compute demand continues to grow exponentially.
預訓練、後訓練和推論，以及現在自主式系統的運算需求持續呈指數成長。 - And now, Vera Rubin. architected for every phase of a genetic AI advancing every pillar of computing, including CPU storage networking and security.
現在是 Vera Rubin。專為自主式 AI 的每一個階段而設計，推進運算的每一個支柱，包括 CPU、儲存、網路和安全性。 - Vera Rubin NVLink 72. 3.6 exa flops of computer. 260 terabytes per second of all to all envy link band.
Vera Rubin NVLink 72。3.6 exaflops 的運算能力。260 TB/s 的 all-to-all NVLink 頻寬。 - The engine supercharging the era of a genetic AI the Vera CPU wrap.
推動自主式 AI 時代的引擎，Vera CPU 機架。 - Designed for orchestration and a genetic workforce the STX rack Ai native storage built with bluefield for. scale out with spectrum X co packaged optics increasing energy efficiency and resilience.
專為協調和自主式工作力而設計，STX 機架，以 BlueField-4 打造的 AI 原生儲存。透過 Spectrum-X 共封裝光學進行 scale-out，提升能源效率和韌性。 - And an incredible new addition.
還有一個令人難以置信的新成員。 - The grog three LP X rack. tightly connected to Vera Rubin rocks LP use massive on chip s RAM a token accelerator to the already incredibly fast.
Groq 3 LPX 機架。與 Vera Rubin 機架緊密連接，LPU 使用大量片上 SRAM，為已經極快的系統提供 token 加速。 - Together 35 times more throughput per megawatt.
合計每百萬瓦的吞吐量提升 35 倍。
Vera Rubin 平台
- The new Vera Rubin platform seven chips five rack scale computers one revolutionary Ai supercomputer for a genetic Ai. 40 million times more compute.
全新的 Vera Rubin 平台，7 種晶片、5 個機架級電腦、一台革命性的 AI 超級電腦，專為自主式 AI 打造。運算量提升 4,000 萬倍。 - In just 10 years.
僅僅 10 年。 - No in the in the good old days when I would say hopper I would hold up a chip.
在過去的美好時光，當我說到 Hopper 的時候我會舉起一顆晶片。 - That’s just adorable.
那真是可愛。 - This is Vera Rubin.
這就是 Vera Rubin。 - When we think Vera Rubin, we think the entire system, vertically integrated, completely with software, extended end-to-end, optimized as one giant system.
當我們想到 Vera Rubin，我們想到的是整個系統，垂直整合，完整包含軟體，端到端延伸，作為一個巨大的系統來最佳化。 - The reason why it’s designed for agentic systems is very clear, because agents, of course, the most important workload is it’s thinking the large language model.
它為什麼是為自主式系統設計的，原因非常清楚，因為對代理來說，最重要的工作負載當然是思考，也就是大型語言模型。 - The large language models are going to get larger and larger and larger.
大型語言模型會變得越來越大、越來越大。 - It’s going to generate more and more tokens more quickly so it can think more quickly, but it also has to access memory.
它會越來越快地生成越來越多的 token，這樣它就能更快地思考，但它也必須存取記憶體。 - It’s going to pound on memory really hard.
它會非常大量地存取記憶體。 - KV cache, structured data, QDF, unstructured data, QVS.
KV cache、結構化資料、QDF、非結構化資料、QVS。 - It’s going to be pounding on the storage system really, really hard, which is the reason why we reinvented the storage system.
它會對儲存系統造成非常、非常大的壓力，這就是我們重新發明儲存系統的原因。 - It is also going to use tools.
它也會使用工具。 - And unlike humans that are more tolerant to slower computers, AI wants the tools to be as fast as possible.
不同於人類對較慢的電腦有更多容忍度，AI 希望工具越快越好。 - These tools, web browsers in the future, they could also be virtual PCs in the cloud.
這些工具，未來的網頁瀏覽器，它們也可以是雲端上的虛擬 PC。 - Those PCs have to be.
那些 PC 必須要。 - And those computers have to be as fast as possible.
那些電腦必須盡可能快。 - We created a brand new CPU.
我們建立了一顆全新的 CPU。 - A brand new CPU that’s designed for extremely high single-threaded performance, incredibly high data output, incredibly good at data processing, and extreme energy efficiency.
一顆全新的 CPU，專為極高的單執行緒效能、極高的資料輸出、極強的資料處理能力和極致的能源效率而設計。 - It is the only data center CPU in the world that uses LPDDR5.
它是全世界唯一使用 LPDDR5 的資料中心 CPU。 - LPDDR5.
LPDDR5。 - And incredible single-thread performance and performance per watt that is unrivaled.
以及無與倫比的單執行緒效能和每瓦效能。 - And so that’s, we built that so that it could go along with the rest of these racks for agentic processing.
所以我們建造它，是為了讓它能與其他機架搭配進行自主式處理。 - And so here it is.
它就在這裡。 - This is the Grace Blackwell, no, Vera Rubin.
這是 Grace Blackwell，不對，Vera Rubin。 - Where is it?
它在哪裡？ - Here it is.
在這裡。 - Okay.
好的。
系統架構與創新
- So this is the Vera Rubin system.
這就是 Vera Rubin 系統。 - Notice, since the last time, 100% liquid cooled.
請注意，相比上次，現在是 100% 液冷。 - All of the cable’s gone.
所有的線纜都不見了。(Ernest: 延伸思考哪些產業可能也被液冷了？) - What used to take.
以前需要。 - What used to take.
以前需要。 - Two days to install now takes two hours.
兩天才能安裝的，現在只需要兩小時。 - Incredible.
令人難以置信。 - And so the manufacturing cycle time is going to dramatically reduce.
所以製造週期時間將會大幅縮短。 - This is also a supercomputer that is cooled by.
這也是一台透過以下方式冷卻的超級電腦。 - It’s cooled by hot water 45 degrees, which takes the pressure off of the data center, takes all of that cost and all of that energy that’s used to cool the data center and makes it available for.
它透過 45 度的熱水冷卻，減輕了資料中心的負擔，將所有用來冷卻資料中心的成本和能源釋放出來，使其可用於。 - The system.
系統。 - This is the secret sauce.
這就是秘密武器。 - It is the only we’re the only company in the world that has today built the 6th 6th generation scale up switching system.
我們是全世界唯一一家已經建造出第 6 代、第 6 代 scale-up 交換系統的公司。 - This is not Ethernet.
這不是乙太網路。 - This is not infinite band.
這不是 InfiniBand。 - This is NVLink.
這是 NVLink。 - This is the 6th generation NVLink.
這是第 6 代 NVLink。 - This is insanely hard to do.
這極其困難。 - Well, it is insanely hard to do period.
它就是極其困難，句點。 - And I’m just super proud of the team NVLink completely legal cool.
我對 NVLink 團隊感到無比自豪，完全合法地酷。 - This.
這個。 - Is the brand new Groq system and I’ll show you a little bit more about it.
是全新的 Groq 系統，我會再多展示一些。 - This system. 8 Groq chips.
這個系統。8 顆 Groq 晶片。 - This is the LP 30 the world’s never seen it anything that the world’s ever seen is V1.
這是 LP30，世界上前所未見的，世界上曾見過的都是第 1 代。 - This is 3rd generation.
這是第 3 代。 - And we’re in volume production now and I’ll show you more about that in just a second.
我們現在已經在量產了，我馬上會展示更多。 - The world’s first.
全世界第一個。 - CPO.
CPO。 - Spectrum X switch.
Spectrum-X 交換器。 - This is also in full production.
這也已經全面量產。 - Co packaged optics.
共封裝光學。 - Optics comes directly onto this chip interfaces directly to silicon electrons gets translated to photons and it gets directly directly connected to this chip.
光學直接到這顆晶片上，直接與矽介面連接，電子轉換成光子，然後直接、直接連接到這顆晶片。 - We invented the process technology with TSMC or the only one in production with it today is called coop.
我們與 TSMC 共同發明了這個製程技術，目前唯一量產的就是所謂的 CoWoS。 - It’s completely revolutionary.
這完全是革命性的。 - Nvidia is in full production with Spectrum X.
NVIDIA 已經全面量產 Spectrum-X。 - This is the Vera system.
這是 Vera 系統。 - Twice the performance per watt of any any CPUs in the world today.
每瓦效能是當今世界上任何 CPU 的兩倍。 - It is also in production.
它也已經在量產。 - Well you know we never we never.
你知道我們從來沒有、從來沒有。 - Thought we would be selling CPU standalone.
想過我們會單獨銷售 CPU。 - We are selling a lot of CPU standalone.
我們正在單獨銷售大量的 CPU。 - This is already for sure going to be a multi billion dollar business for us.
這對我們來說肯定已經會是一項數十億美元的業務。 - So I’m very very pleased with our CPU architects we’ve designed a revolutionary CPU.
所以我對我們的 CPU 架構師非常、非常滿意，我們設計了一顆革命性的 CPU。 - And this.
而這。 - Is the.
是。 - ConnectX-9.
ConnectX-9。 - Powered with Vera CPU the bluefield for STX our new storage platform.
搭載 Vera CPU、BlueField-4 的 STX，我們全新的儲存平台。 - Okay so these are the four these are the the racks and it’s connected.
好的，這就是那四個，這些就是那些機架，而且它們是連接在一起的。
3️⃣ AI 工廠：Vera Rubin
NVLink 與乙太網路機架
- Each one of these racks the NVLink rack.
這些機架中的每一個都是 NVLink 機架。 - This is.
就是這個。 - I’ve shown you guys this before it’s a super heavy and seems to get heavier every year.
我之前給你們展示過，它超級重，而且每年似乎都變得更重。 - Because I think there’s just more cables in there every year.
因為我覺得每年裡面的線纜都越來越多。 - And so so this is the NVLink rack we’ve also taken this technology because it is so.
所以這就是 NVLink 機架，我們也採用了這項技術，因為它非常。 - Efficient to create a data center with these cabling systems structured cables so we decided to do that for Ethernet so this is Ethernet two hundred fifty six.
用這些結構化佈線系統來建構資料中心非常有效率，所以我們決定把同樣的做法用在乙太網路上，這就是乙太網路 256。 - Liquid cooled nodes in one rack and it is also connected with.
一個機架中有液冷節點，而且還連接了。 - These incredible connectors.
這些令人驚豔的連接器。 - You guys want to see.
你們想看看嗎。
Rubin Ultra 與 Kyber 機架系統
- Reuben ultra.
Rubin Ultra。 - So this is the Reuben ultra compute node.
這就是 Rubin Ultra 運算節點。 - Unlike.
不像。 - Reuben.
Rubin。 - That slides in horizontally.
那是水平滑入的。 - Reuben. in Ultra goes into a whole new rack, it’s called Kyber, that enables us to connect 144 GPUs in one MV-Link domain.
Rubin Ultra 則安裝到一個全新的機架中，它叫做 Kyber，讓我們能夠在一個 NVLink 域中連接 144 顆 GPU。 - And so the Kyber rack, this, I could lift it, I’m sure, but I won’t.
所以 Kyber 機架，這個嘛，我確定我能舉起來，但我不會這樣做。 - It’s quite heavy.
它相當重。 - This is one compute node and it slides into the Kyber rack vertically.
這是一個運算節點，它垂直滑入 Kyber 機架中。 - This is where it connects into.
這是它連接的位置。 - This is the mid-plane.
這是中間板。 - The Kyber racks, those four top MV-Link connectors slide in and connect into this and this becomes one of the nodes.
Kyber 機架頂部的四個 NVLink 連接器滑入並連接到這裡，然後這就成為其中一個節點。 - And each one of these racks is a different compute node and this is the amazing part.
而每一個這樣的機架都是不同的運算節點，這是最驚人的部分。 - This is the mid-plane and the back of the mid-plane, instead of the cabling system, which has its limit.
這是中間板以及中間板的背面，取代了有其限制的佈線系統。 - In terms of how far we could drive cables, copper cables, we now have this system to connect 144 GPUs.
就銅纜能傳輸的距離而言，我們現在有了這套系統來連接 144 顆 GPU。 - This is the new MV-Link.
這就是新的 NVLink。 - This sits also vertically and it connects into the mid-planes on the back.
它也是垂直放置，連接到背面的中間板。 - Compute in the front, MV-Link switches in the back, one giant computer.
前方是運算，後方是 NVLink 交換器，組成一台巨型電腦。 - Okay?
好嗎？ - So that is.
所以那就是。 - Ruben Ultra.
Rubin Ultra。 - As I mentioned, as I mentioned, how about we take this back down?
正如我提到的，正如我提到的，我們把這個放下來如何？ - I need the rest of my slides.
我需要我剩下的投影片。 - Oh, it’s coming down?
噢，它要降下來了？ - Okay.
好的。 - Thank you, Janine.
謝謝你，Janine。 - This is what happens when you, this is what happens when you don’t practice.
這就是當你，這就是當你沒有彩排時會發生的事。 - Okay, all right.
好的，好吧。 - So, you saw, you, take your time, just don’t get hurt.
所以，你們看到了，你，慢慢來，別受傷就好。
Token 工廠經濟學
AI 工廠的吞吐量與 Token 速度
- You saw, you saw this slide.
你們看到了，你們看到了這張投影片。 - You know, only on NVIDIA’s keynote will you see last year’s slide presented again.
你知道，只有在 NVIDIA 的主題演講中你才會看到去年的投影片再次出現。 - And the reason for that is I just want to let you know that last year I told you something very, very important and it’s so important it’s worthwhile to tell you again.
原因是我只是想讓你們知道，去年我告訴了你們一些非常非常重要的事情，而且它重要到值得再跟你們說一次。 - This is probably the single most important chart for the future of AI factories.
這可能是對 AI 工廠未來最重要的一張圖表。 - And every CEO, every CEO in the world will be tracking it, will be studying it very deeply.
而且全世界每一位 CEO，每一位 CEO 都會追蹤它，都會非常深入地研究它。 - It’s much, much more complicated than this.
它比這複雜得多。 - It’s multi-dimensional.
它是多維度的。 - But you will be studying the throughput and the token speed of your AI factories.
但你們會研究你們 AI 工廠的吞吐量和 token 速度。 - The throughput, token speed at ISO power, because that’s all the power you have.
在相同功耗下的吞吐量和 token 速度，因為那就是你所有的電力。
Token 定價與市場區隔
- And that analysis is going to lead directly to your revenues.
而這個分析將直接導向你們的營收。 - What you do this year will show up precisely next year as your revenues.
你們今年做的事情將精確地在明年反映為你們的營收。 - And this chart is what it’s all about.
而這張圖表就是一切的核心。 - And I said on the vertical axis, on the vertical axis, thank you guys, on the vertical axis is throughput.
我說過在縱軸上，在縱軸上，謝謝你們，縱軸是吞吐量。 - On the horizontal axis is token rate.
橫軸是 token 速率。 - Today I’m going to show you this.
今天我要給你們看這個。 - Because we’re able because we’re now able to increase the token speed and because model sizes are increasing.
因為我們能夠，因為我們現在能夠提高 token 速度，而且模型大小也在增加。 - Because the token length the context length depending on the different grades of different application use case continues to grow from maybe a hundred thousand tokens input length to maybe millions.
因為 token 長度，也就是上下文長度，取決於不同應用場景的不同等級，持續從大約 10 萬個 token 的輸入長度成長到可能數百萬個。 - The token input length is growing and also the output token length is growing and so all of these play into ultimately.
Token 輸入長度在增長，輸出 token 長度也在增長，所以這一切最終都會影響到。 - The marketing and the pricing of future tokens tokens are the new commodity.
未來 token 的行銷和定價，token 是新的商品。 - And like all commodities once it reaches an inflection once it becomes mature or becomes maturing it will segment into different parts.
就像所有商品一樣，一旦它達到轉折點，一旦它變得成熟或正在成熟，它就會區隔成不同的部分。 - The high throughput.
高吞吐量。 - Low speed could be used for the free tier.
低速可以用於免費層。 - The next year could be the medium tier larger model maybe higher speed for sure larger input context length.
下一層可能是中等層級，更大的模型，也許更高的速度，肯定更大的輸入上下文長度。 - That translates to a different price point you could see from all the different services this one is free it’s a free tier.
這轉化為不同的價格點，你可以從所有不同的服務中看到，這個是免費的，是免費層。 - The first year could be $3 per million tokens the next year could be $6 per million tokens you would like to be able to keep pushing this boundary.
第一層可能是每百萬 token 3 美元，下一層可能是每百萬 token 6 美元，你會希望能夠持續推進這個邊界。 - Because.
因為。 - The.
那個。 - Low speed.
低速。 - Speed.
速度。 - The larger the model smarter the more input token context length more relevant.
模型越大越聰明，輸入 token 上下文長度越多就越相關。 - The higher the speed.
速度越高。 - The long the more you can think and iterate smarter AI models so this is about smarter AI models.
越長就越能思考和迭代出更聰明的 AI 模型，所以這是關於更聰明的 AI 模型。 - And when you have smarter and models each one of these clicks allows you to increase the price so this is $45.
當你有更聰明的模型時，每一個層級的提升都讓你可以提高價格，所以這是 45 美元。 - And maybe one day there’ll be a premium model that allows you a premium service that allows you to.
也許有一天會有一個高級模型，讓你提供一個高級服務，讓你能夠。 - Generate token speeds that are incredibly high because you’re in a critical path or maybe you’re doing really long research and $150 per million tokens is just not a thing so let’s translate that.
產生極高的 token 速度，因為你在關鍵路徑上，或者也許你正在做非常長時間的研究，而每百萬 token 150 美元根本不算什麼，所以讓我們來換算一下。 - Suppose you were to use 50 million tokens per day as a researcher at $150 per million tokens.
假設你作為研究人員每天使用 5000 萬個 token，以每百萬 token 150 美元計算。 - As it turns out as a research team that’s not even a thing.
事實證明，對於一個研究團隊來說，這根本不算什麼。 - So we believe that this is the future this is where I wants to go this is where it is today.
所以我們相信這就是未來，這就是 AI 要去的方向，這就是它今天所在的位置。
各層級效能提升
- It had to start here to establish the value and establishes usefulness and get better and better and better.
它必須從這裡開始，建立價值並確立實用性，然後變得越來越好。 - In the future you’re going to see most services encompass it encompass all of that this is hopper.
在未來你會看到大多數服務涵蓋它，涵蓋所有這些，這是 Hopper。 - Hoppers started and I moved it moved the chart this is 50 this is 100.
Hopper 開始了，然後我移動了，移動了圖表，這是 50，這是 100。 - Hopper looks like this and you would have expected hopper the next generation to be higher but nobody would have expected it to be that much higher this is Grace Blackwell what Grace Blackwell is.
Hopper 看起來像這樣，你會預期下一代 Hopper 會更高，但沒有人會預期它會高出那麼多，這就是 Grace Blackwell，Grace Blackwell 做到的是。 - Blackwell did is at your free tier increase your throughput tremendously.
Blackwell 做到的是在你的免費層大幅提升你的吞吐量。 - However.
然而。 - Where you mostly monetize your service it increase your throughput by 35 times.
在你主要獲利的服務層級，它將你的吞吐量提升了 35 倍。 - This is no different than any product that every company makes the higher the tier the higher the quality the higher the performance the lower the volume the lower the capacity.
這與每家公司製造的任何產品沒有什麼不同，層級越高，品質越高，效能越高，產量越低，容量越低。 - And so it is no different than any other business in the world and so.
所以這與世界上任何其他業務沒有什麼不同，所以。 - Now.
現在。 - We’re able to increase.
我們能夠提升。 - This tier by 35 X and we introduced a whole new tier.
這個層級 35 倍，而且我們引入了一個全新的層級。 - This this is the benefit of Grace Blackwell a huge jump over hopper.
這就是 Grace Blackwell 的好處，相較於 Hopper 的巨大躍升。 - Well this is what we’re doing with.
好的，這就是我們正在做的。 - Okay so this is Grace Blackwell.
好的，所以這是 Grace Blackwell。 - Okay let me just reset reset this.
好的，讓我重新設定一下。 - And this is very ruby.
而這是 Vera Rubin。 - Okay.
好的。 - Now just think just think what just happened.
現在想想，想想剛剛發生了什麼。 - At every single tier at every single tier and every single tier we increase the throughput.
在每一個層級，每一個層級，每一個層級，我們都提升了吞吐量。 - And at the tier that where your highest ASP and your most valuable segment we increased it by 10 X.
而在你最高 ASP 和最有價值的區隔中，我們提升了 10 倍。 - That is the hard work this is incredibly hard to do out here this is the benefit of every link 72 this is the benefit of.
這就是艱苦的工作，在這裡做到這些是難以置信地困難，這就是 NVLink 72 的好處，這就是。 - This is the benefit of extremely low latency this is the benefit of extreme co-design that we can shift the entire area.
這就是極低延遲的好處，這就是極致協同設計的好處，讓我們能夠移動整個區域。 - Now what does it mean from a customer perspective in the end suppose I were to take all of that and I just you know multiply it against.
現在從客戶的角度來看這意味著什麼呢，最終假設我把所有這些加起來，然後你知道，乘以。 - Suppose I took 25% of my power used it in free tier 25% of my power in the medium tier 25% of my power in the high tier and 25% of my power in the premium tier my data center only has a gigawatt.
假設我把 25% 的電力用在免費層，25% 的電力用在中等層，25% 的電力用在高階層，25% 的電力用在高級層，我的資料中心只有 1 GW。 - And so I get to decide.
所以我可以決定。 - How I want to distribute the free tier allows me to track more customers.
我要如何分配，免費層讓我能夠吸引更多客戶。 - This allows me to serve my most valuable customers.
這讓我能夠服務我最有價值的客戶。 - And the combination the product of all that allows you basically your revenues the revenues you can generate assuming this simplistic example allows Blackwell to generate 5 times more revenues.
而這些的組合，所有這些的乘積基本上就是你的營收，假設這個簡化的例子，你能產生的營收讓 Blackwell 能夠產生 5 倍的營收。 - Very ruben to generate 5 times.
Vera Rubin 再產生 5 倍。 - Yeah.
是的。 - So Vera Rubin you should get there as soon as you can.
所以 Vera Rubin，你們應該盡快取得。 - And the reason for that is because your cut your cost of tokens goes down and your throughput goes up now but we want even more we want even more and so let me just show you back to this.
原因是因為你的 token 成本下降了，你的吞吐量也提升了，但我們還想要更多，我們還想要更多，所以讓我帶你們回到這個。
吞吐量與延遲的挑戰
- This is.
這是。 - As you as I as I told you this throughput requires a ton of flops this latency.
正如我跟你們說的，這個吞吐量需要大量的浮點運算，而這個延遲。 - This interactivity.
這種互動性。 - Requires enormous amount of bandwidth computers don’t like extreme amount of flops extreme amount of bandwidth because there’s only so much.
需要巨大的頻寬，電腦不喜歡極端的浮點運算量和極端的頻寬量，因為只有這麼多。 - Surface area for chips that any systems has and so optimizing for high throughput and optimizing for low latency are in fact enemies of each other.
任何系統的晶片面積都是有限的，所以最佳化高吞吐量和最佳化低延遲實際上是彼此的敵人。 - And so this is what happened when we combined with Groq okay and so we we acquired the team that worked on the Groq chips and license the technology and we’ve been working together now.
所以這就是當我們與 Groq 結合時發生的事情，好的，所以我們收購了研發 Groq 晶片的團隊並授權了技術，而且我們現在已經在一起合作了。 - To integrate the system.
來整合這個系統。 - This is what that looks like.
這就是它的樣子。
Groq 整合提升吞吐量
- So at the most valuable tier at the most valuable tier we’re now going to increase performance by 35 X.
所以在最有價值的層級，在最有價值的層級，我們現在要將效能提升 35 倍。 - Now this.
現在這個。 - Very simple chart revealed to you exactly the reason why Nvidia is.
非常簡單的圖表向你們揭示了 NVIDIA 為什麼。 - So strong in the vast majority of the workloads so far and the reason for that is because up in this area.
在到目前為止絕大多數的工作負載中如此強大的確切原因，原因是因為在這個區域。 - Throughput matters so much.
吞吐量非常重要。 - You’re not going to be able to see this.
你們可能看不到這個。 - But if you look at the chart that we’ve just shown you.
但如果你看一下我們剛剛展示的圖表。 - You can see that NVLink 72 is so game changing it is exactly the right architecture and it’s even hard to beat even as you add Groq to it.
你可以看到 NVLink 72 是如此地具有顛覆性，它正是正確的架構，即使加上 Groq 也很難超越它。 - However.
然而。 - If you extended this chart way out here and you said you wanted to have services that delivers not 400 tokens per second but 1000 tokens per second.
如果你把這張圖表延伸到這裡，你說你想要的服務不是每秒 400 個 token，而是每秒 1000 個 token。 - All of a sudden.
突然之間。 - Envying 72 runs out of steam.
NVLink 72 就力不從心了。 - And.
而且。 - Simply.
根本。 - Can’t get there.
無法達到。 - And this is what happens.
而這就是會發生的事。 - When we push that out.
當我們把它推出去的時候。 - So it goes out beyond.
所以它超越了。 - Thank you.
謝謝。 - Goes out beyond even the limits of what any link 72 can do and if you were to do that.
超越了 NVLink 72 所能做到的極限，而如果你這樣做的話。 - Translate that into revenues.
把它轉化為營收。 - Relative to Blackwell.
相對於 Blackwell。 - Vera Rubin is 5 X.
Vera Rubin 是 5 倍。 - If most of your workload is high throughput.
如果你的大部分工作負載是高吞吐量的。 - I would stick with just 100 percent.
我會堅持 100% 使用。
為程式設計與工程任務最佳化
- Vera Rubin.
Vera Rubin。 - If a lot of your workload wants to be.
如果你的很多工作負載想要是。 - Coding and very high valued engineering token generation I would add Groq to it.
程式設計和非常高價值的工程 token 生成，我會加入 Groq。 - I would add Groq to maybe 25% of my total data center the rest of my data center is all 100% Vera Rubin.
我會在大約 25% 的資料中心加入 Groq，其餘的資料中心全部 100% 使用 Vera Rubin。 - And so that gives you a sense of how you would add.
所以這讓你了解你會如何加入。 - Grog to Vera Rubin and extend its performance and extend its value even more.
Groq 到 Vera Rubin 中，並進一步擴展其效能和價值。 - This is what happens.
這就是會發生的事。
Groq 架構與解耦推論
Groq 的確定性資料流架構
- Very.
非常。 - This is a contrast the reason why the reason why Groq was so attractive to me.
這是一個對比，Groq 對我如此有吸引力的原因。 - Is because their computing system a deterministic data flow processor it is statically compiled it is compiler scheduled.
是因為他們的運算系統是一個確定性資料流處理器，它是靜態編譯的，是由編譯器排程的。 - Meaning the compiler figures out when the date when to do the compute the computing data arrives at the same time all of that is done statically in advance.
這意味著編譯器預先算出何時進行運算，運算資料同時到達，所有這些都是預先靜態完成的。 - And scheduled.
並且排程好。 - Completely in software.
完全透過軟體。 - There’s no dynamic scheduling.
沒有動態排程。 - The architecture is designed with massive amounts of SRAM.
該架構設計了大量的 SRAM。 - It is designed just for inference.
它專為推論而設計。 - This one workload now this one workload as it turns out is the workload of AI factories.
這一個工作負載，而這一個工作負載事實證明就是 AI 工廠的工作負載。 - And as the world continues to increase the amount of high speed tokens and wants to generate with super smart tokens and wants to generate the value of disintegration is going to get even higher.
隨著全世界持續增加高速 token 的數量，並且想要用超聰明的 token 來生成，解耦的價值將會變得更高。 - And so these are two.
所以這是兩個。 - Extreme processors you could see.
你可以看到的極端處理器。
以 Dynamo 解耦推論
- One chip 500 megabytes.
一顆晶片 500 MB。 - One there are Rubin chip one Rubin chip 288 gigabytes.
一顆 Rubin 晶片，一顆 Rubin 晶片 288 GB。 - It would take a lot of.
需要很多。 - Groq chips to be able to hold the parameter size of Rubin as well as all of the context that has to go the KV cash that has to go along with it.
Groq 晶片才能夠容納 Rubin 的參數規模以及所有必須伴隨的上下文和 KV 快取。 - So that limited Groqs ability to really reach the mainstream to really take off.
所以那限制了 Groq 真正進入主流、真正起飛的能力。 - Until we had a great idea.
直到我們有了一個好主意。 - What if we disaggregated inference altogether with a piece of software called Dynamo.
如果我們用一套叫做 Dynamo 的軟體來完全解耦推論呢。 - What every re architected the way that inferences done in the pipeline so that we could put the work that makes perfect sense on Rubin.
重新架構推論在管線中的執行方式，這樣我們就可以把完全合理的工作放在 Rubin 上。 - And then offload the decode generation the low latency the bandwidth limited challenged part of the workload for Groq and so we.
然後將解碼生成，也就是低延遲、頻寬受限的那部分工作負載卸載到 Groq 上，所以我們。 - United unified.
統一了。 - Two processors of extreme differences one for high throughput one for low latency.
兩個極端差異的處理器，一個用於高吞吐量，一個用於低延遲。 - It still doesn’t change the fact that we need a lot of memory and so Groq we’re just going to add a whole bunch of Groq chips.
這仍然不能改變我們需要大量記憶體的事實，所以 Groq，我們就是要加入一大堆 Groq 晶片。
統一處理器架構與量產
- Which expands the amount of memory it has and so if you could just imagine.
這擴展了它的記憶體容量，所以如果你能想像一下。 - Out of a trillion parameter model we have to store all of that in Groq chips however it sits next to.
對於一個兆級參數的模型，我們必須把所有這些儲存在 Groq 晶片中，但它就在旁邊。 - NVIDIA Vera Rubin.
NVIDIA Vera Rubin。 - Where.
在那裡。 - We could we could hold the massive amounts of KB cash that’s necessary in processing all of these agentic AI systems.
我們可以容納處理所有這些自主式 AI 系統所需的大量 KV 快取。 - It’s based upon this idea of this aggregated inference we do the pre fill that’s the easy part but we also tightly integrate the decode so.
它基於這個解耦推論的想法，我們執行預填充，那是簡單的部分，但我們也緊密整合了解碼，所以。 - The attention part of decode is done on NVIDIA’s Vera Rubin which needs a lot of math and the.
解碼的注意力部分在 NVIDIA 的 Vera Rubin 上完成，那需要大量的運算，而。 - Feed forward network part of it.
前饋網路的部分。 - The decode part is done, the token generation part is done on Vera Rubin, on the Groq chip.
解碼部分完成了，token 生成部分在 Groq 晶片上完成。 - The two of them working tightly coupled together over today, Ethernet with a special mode to reduce its latency by about half.
兩者透過乙太網路緊密耦合在一起，使用特殊模式將延遲降低約一半。 - And so that capability allows us to integrate these two systems.
所以這個能力讓我們能夠整合這兩個系統。 - We run Dynamo, this incredible operating system for AI factories on top of it.
我們在上面執行 Dynamo，這個為 AI 工廠打造的不可思議的作業系統。 - And you get 35 times increase. 35 times increase, not to mention additional new tiers of inference performance for token generation the world’s never seen.
然後你獲得 35 倍的提升。35 倍的提升，更不用說世界從未見過的 token 生成推論效能的額外新層級。 - So this is it.
就是這樣。 - This is Groq.
這就是 Groq。 - The Vera Rubin systems, including Groq, I want to thank Samsung, who manufactures the Groq LP30 chip for us, and they’re cranking us hard as they can.
Vera Rubin 系統，包括 Groq，我想感謝 Samsung，他們為我們製造 Groq LP30 晶片，他們正在盡全力加速生產。 - I really appreciate you guys.
我真的很感謝你們。 - We’re in production with the Groq chip, and, you know, we’ll ship it in the second half, probably about Q3 timeframe.
我們已經在量產 Groq 晶片，而且，你知道，我們會在下半年出貨，大概是第三季的時間。
Vera Rubin 系統元件
Vera Rubin 系統與 CPU
- Okay?
好嗎？ - Groq LPX.
Groq LPX。 - Vera Rubin, you know, it’s kind of hard to imagine any more customers.
Vera Rubin，你知道，很難想像還能有更多客戶。 - You know?
你知道嗎？ - And, you know, I’m not going to say that.
而且，你知道，我不打算這樣說。 - I’m not going to say that.
我不打算這樣說。 - And the really great thing is Grace Blackwell’s early sampling of it was really complicated because of coming together with NVLink 72.
而真正棒的是 Grace Blackwell 的早期樣品非常複雜，因為要與 NVLink 72 整合在一起。 - But the sampling of Vera Rubin is just going incredibly well.
但 Vera Rubin 的樣品進展非常順利。 - And in fact, Satya, I think, texted out already that the first Vera Rubin rack is already up and running at Microsoft Azure.
事實上，我想 Satya 已經發推文說第一個 Vera Rubin 機架已經在 Microsoft Azure 上啟動並執行了。 - And so I’m super excited for them.
所以我為他們感到非常興奮。 - We’re going to keep cranking these things out.
我們會持續大量生產這些東西。 - We have now set up a supply chain that can manufacture thousands a week.
我們現在已經建立了一條每週可以生產數千台的供應鏈。 - And we have a very big chunk of these systems essentially multi gigawatts of AI factories per month inside our supply chain.
而且我們的供應鏈中有非常大量的這些系統，基本上每月相當於數 GW 的 AI 工廠。 - And so we’re going to crank out these Vera Rubin racks while we’re cranking out the GB300 racks.
所以我們會在生產 GB300 機架的同時持續生產 Vera Rubin 機架。 - We are in full production.
我們已經全面量產。 - The Vera CPUs, incredibly successful.
Vera CPU，非常成功。 - And the reason for that is because AI needs CPUs for tool use.
原因是 AI 需要 CPU 來做工具使用。 - And Vera CPU was designed just perfectly for that sweet spot.
而 Vera CPU 正是為那個甜蜜點完美設計的。 - Incredible.
令人驚嘆。 - For the next generation of data processing.
對於下一代資料處理而言。 - Vera CPU is ideal.
Vera CPU 是理想的選擇。 - The Vera CPU plus CX9 connected into the Bluefield 4 stack.
Vera CPU 加上 CX9 連接到 BlueField 4 堆疊中。
儲存產業採用與 KV 快取
- 100% of the world’s storage industry is joining us on this system.
全球 100% 的儲存產業都加入了我們這個系統。 - And the reason for that is because they see exactly the same thing.
原因是因為他們看到了完全相同的事情。 - The storage system is going to get pounded.
儲存系統將會承受巨大的壓力。 - It’s going to get pounded because we used to have humans using the storage systems.
它會承受巨大壓力，因為我們過去是人類在使用儲存系統。 - We used to have humans using sequel.
我們過去是人類在使用 SQL。 - Now we’re going to have a eyes using the storage systems and it’s going to store.
現在我們將有 AI 在使用儲存系統，而且它要儲存。 - Who DF accelerated storage?
誰需要 GPU 加速儲存？ - Who vs accelerated storage as well as very importantly KV caching.
誰需要加速儲存以及非常重要的 KV 快取。 - OK, so this is the Vera Rubin system.
好的，這就是 Vera Rubin 系統。
效能躍升與發展藍圖
- Now what’s amazing is this?
現在令人驚嘆的是什麼呢？ - In just two years time.
在僅僅兩年的時間內。 - And a one gigawatt factory.
在一座 1 GW 的工廠中。 - In just two years time in one gigawatt factory.
在僅僅兩年內，在一座 1 GW 的工廠中。 - Using.
使用。 - Using the mathematics that I showed you earlier.
使用我之前展示給你們的數學計算。 - We’re as Moore’s law would have given us a couple of steps we would have you know.
按照摩爾定律本來會給我們的幾個進步步驟，我們會。 - X factored.
倍增。 - The number of transistors we would X factor the number of flops.
電晶體數量，我們會倍增浮點運算數量。 - We were to X factored the number of.
我們會倍增。 - Amount of bandwidth but with this architecture we’re going to take our token generation speed token generation rate.
頻寬量，但透過這個架構，我們將把我們的 token 生成速度、token 生成速率。 - From. 22 million.
從 2200 萬。 - To 700 million.
提升到 7 億。 - Three hundred and. 50 times increase.
350 倍的提升。 - This is this is the power of extreme co-design this is what I mean when we integrate and optimize vertically but then we open it horizontally for everybody to enjoy this is our road map.
這就是極致協同設計的力量，這就是我所說的我們垂直整合和最佳化，然後水平開放讓每個人都能享受，這就是我們的發展藍圖。 - Very quickly.
非常快速。
硬體發展藍圖：從 Rubin Ultra 到 Feynman
未來架構與升級
- Blackwell is here.
Blackwell 在這裡。 - The Oberon system.
Oberon 系統。 - In the case of Rubin.
在 Rubin 的情況下。 - We have the Oberon system.
我們有 Oberon 系統。 - We’re always backwards.
我們始終向下。 - And we’re always going to be compatible so that if you wanted to not change anything and just keep on moving through with the new architecture you could do so.
而且我們始終會保持相容性，這樣如果你不想改變任何東西，只想繼續使用新架構，你就可以這樣做。 - The old system the standard rack system Oberon still available Oberon is copper scale up and with Oberon.
舊系統、標準機架系統 Oberon 仍然可用，Oberon 是銅纜擴展，而且透過 Oberon。 - We could also use optical scale out or excuse me optical scale up.
我們也可以使用光學橫向擴展，不好意思，是光學縱向擴展。 - To expand to NVLink 576.
來擴展到 NVLink 576。 - Okay.
好的。 - And so there’s a lot of conversation about.
所以有很多關於這方面的討論。 - Is.
是否。 - Media going to copper scale up or optical scale up we’re going to do both.
NVIDIA 會用銅纜擴展還是光學擴展，我們兩者都要做。 - So we’re going to have NVLink 144 with Kyber.
所以我們會有 Kyber 的 NVLink 144。 - And then with up to Ron.
然後透過 Oberon。 - After on Oberon we’re going to NVLink 72 plus.
在 Oberon 之後我們會有 NVLink 72 加上。 - Optical to get to NVLink 576.
光學來達到 NVLink 576。 - The next generation of.
下一代。 - Rubin with Rubin ultra we have the Rubin ultra chip which is coming which is.
Rubin，透過 Rubin Ultra，我們有即將推出的 Rubin Ultra 晶片，它正在。 - Taping out.
進行 Tape out。 - And we have a brand new chip LP 35.
而且我們有一顆全新的晶片 LP 35。 - LP 35 will for the first time incorporate.
LP 35 將首次整合。 - NVIDIA’s.
NVIDIA 的。 - MVFP for computing structure.
NVFP 運算架構。 - Give you another few X X factor speed up.
給你再多幾倍的速度提升。 - Okay and so this is.
好的，所以這是。 - Oberon NVLink 72.
Oberon NVLink 72。
下一代與新 CPU
- Optical scale up.
光學縱向擴展。 - And.
和。 - It uses spectrum 6 the world’s first co packaged.
它使用 Spectrum 6，全球第一個共封裝的。 - Optical.
光學。 - And.
和。 - All of this is in production.
所有這些都在量產中。 - The next generation.
下一代。 - From here.
從這裡開始。 - Is.
是。 - Fineman.
Feynman。 - Fineman has a new GPU, of course.
Feynman 當然有一顆新的 GPU。 - It also has a.
它也有一個。 - New LP you.
新的 LP。 - LP 40.
LP 40。 - Big step up.
大幅提升。 - Incredible, incredible new technology.
令人驚嘆的全新技術。 - Now, uniting.
現在，結合。 - The scale of NVIDIA and.
NVIDIA 的規模和。 - The Groq team building together, LP 40.
Groq 團隊一起打造，LP 40。 - It’s going to be incredible.
它將會令人驚嘆。 - A brand new CPU called Rosa.
一顆全新的 CPU 叫做 Rosa。
連接性與擴展方案
- A.
一個。 - Short for Roslyn.
Roslyn 的簡稱。 - Bluefield 5, which connects the next CPU with the next.
BlueField 5，它連接下一代 CPU 和下一代。 - Super neck.
SuperNIC。 - CX 10.
CX10。 - We will have Kyber.
我們將有 Kyber。 - Which is copper scale up.
也就是銅纜縱向擴展。 - We will also have Kyber.
我們也將有 Kyber。 - CPO scale up.
CPO 縱向擴展。 - So, for the first time.
所以，史上第一次。 - We will scale up.
我們將縱向擴展。 - With both.
同時使用。 - Copper.
銅纜。 - And.
和。 - Copackage optics.
共封裝光學。 - Okay?
好嗎？ - And so.
所以。
產能擴充與生態系成長
- A lot of people have been asking.
很多人一直在問。 - You know, Jensen, are is copper going to.
你知道，Jensen，銅纜是否還會。 - Still be important.
仍然重要。 - The answer is yes.
答案是肯定的。 - Jensen, are you going to scale up.
Jensen，你們是否會縱向擴展。 - Optical.
光學。 - Yes.
是的。 - Are you going to scale out optical.
你們是否會橫向擴展光學。 - Yes.
是的。 - And so, for everybody who is in our ecosystem.
所以，對於我們生態系中的每一個人。 - We need a lot more capacity.
我們需要多很多的產能。 - And that’s really.
而那真的是。 - The key.
關鍵。 - We need a lot more capacity for copper.
我們需要多很多的銅纜產能。 - We need a lot more capacity for optics.
我們需要多很多的光學產能。 - We need a lot more capacity for CPO.
我們需要多很多的 CPO 產能。 - And that’s the reason why we’ve been working with all of you.
這就是為什麼我們一直與你們所有人合作。 - To lay the foundation for this level of growth.
為這個層級的成長奠定基礎。 - And so.
所以。 - We’ll have all of that.
我們會擁有所有這些。
NVIDIA 轉型為 AI 基礎設施公司
- Let me see if I missed everything.
讓我看看是否有遺漏。 - That’s it.
就是這樣。 - Every single year.
每一年。 - Brand new architecture.
全新架構。 - Very quick.
非常快速。 - Very quickly.
非常快速地。 - We’re going to.
我們將要。 - We’re going to.
我們將要。 - Very quickly.
非常快速地。 - NVIDIA went from a chip company.
NVIDIA 從一家晶片公司。 - To a.
變成一家。 - A factory company or infrastructure company.
工廠公司或基礎設施公司。 - AI computing company.
AI 運算公司。 - These systems.
這些系統。 - And now.
而現在。
4️⃣ DSX 與 Omniverse
AI 工廠設計與 Omniverse
- We’re building entire AI factories.
我們正在建造完整的 AI 工廠。 - There’s so much power.
有這麼多的電力。 - That is squandered in these AI factories.
在這些 AI 工廠中被浪費掉了。 - We want to make sure that these AI factories come together.
我們希望確保這些 AI 工廠能夠整合在一起。 - Design in the best possible way.
以最佳方式進行設計。 - Most of these components never meet each other.
這些元件大多數從未彼此接觸過。 - Most of us technology vendors.
我們大多數技術供應商。 - Now we all know each other.
現在我們都認識彼此了。 - But in the past.
但在過去。 - We never met each other until the data center.
我們直到資料中心才見到彼此。 - That can’t happen.
這種情況不能再發生了。 - We’re building super complex systems.
我們正在建造超級複雜的系統。 - And so we have to meet each other virtually somewhere else.
所以我們必須在其他地方以虛擬方式見面。 - And so we created Omniverse.
因此我們建立了 Omniverse。 - And the Omniverse DSX world.
以及 Omniverse DSX 的世界。 - A platform.
一個平台。 - Where all of us can meet.
讓我們所有人都能在此會合。 - And design these giga factories.
並設計這些超大規模工廠。 - You know, gigawatt AI factories.
也就是十億瓦級的 AI 工廠。 - Virtually.
以虛擬方式。 - In system.
在系統中。 - We have simulation.
我們有模擬功能。 - Systems for the racks, for mechanical.
針對機架、機械方面的系統。 - Thermal.
散熱。 - Electrical.
電力。 - Networking.
網路。 - Those simulation systems integrated into.
這些模擬系統整合到。 - All of our ecosystem partners of incredible tools companies.
我們所有生態系合作夥伴的優秀工具公司中。 - We also operated.
我們也進行了運作。 - Connected to the grid.
連接到電網。 - So that we could.
這樣我們就能。 - Interact with each other.
彼此互動。 - Send each other.
互相傳送。 - Information.
資訊。 - So that we could adjust.
這樣我們就能調整。 - Grid power.
電網電力。 - And data center power.
以及資料中心電力。 - Accordingly.
做出相應調整。 - Saving energy.
節省能源。 - And so.
因此。
DSX 平台元件與 API
- We could.
我們可以。 - And then.
接著。 - Inside the data center.
在資料中心內部。 - Using Max Q.
使用 Max-Q。 - So that we could adjust the system dynamically across power and cooling and all of the different technologies we all work on together.
這樣我們就能在電力、散熱以及我們共同合作的所有不同技術之間動態調整系統。 - So that we.
這樣我們。 - Leave no power squandered.
不浪費任何電力。 - So that.
這樣。 - We run at the most optimal rate.
我們以最佳效率執行。 - To deliver enormous amount of.
以產出大量的。 - Token throughput.
Token 吞吐量。 - There’s no question in my mind.
我毫不懷疑。 - There’s a factor of two in here.
這裡有兩倍的差距。 - And a factor of two at the scale we’re talking about.
在我們討論的規模下，兩倍的差距。 - Is gigantic.
是非常巨大的。 - We call this the NVIDIA DSX platform.
我們稱之為 NVIDIA DSX 平台。 - And just as all of our platforms.
就像我們所有的平台一樣。 - There’s the hardware layer.
有硬體層。 - There’s the library layer.
有函式庫層。 - And there’s the ecosystem layer.
還有生態系層。 - It’s exactly the same way.
完全是相同的架構。 - Let’s show it to you.
讓我們展示給你看。 - The greatest infrastructure build out in history is underway.
史上最大規模的基礎設施建設正在進行中。 - The world is racing to build chip system and AI factories.
全世界正在競相建造晶片系統和 AI 工廠。 - And every month of delay costs billions in lost revenues.
每延遲一個月就會造成數十億美元的營收損失。 - AI factory revenues are equal to tokens per watt.
AI 工廠的營收等於每瓦特的 token 產出。 - So with power constraints.
因此在電力受限的情況下。 - Every unused watt is revenue lost.
每一瓦未使用的電力都是損失的營收。 - NVIDIA DSX is an Omniverse digital twin blueprint.
NVIDIA DSX 是一個 Omniverse 數位孿生藍圖。 - For designing and operating AI factories for maximum token throughput.
用於設計和營運 AI 工廠，以實現最大 token 吞吐量。 - Resilience and energy efficiency.
韌性和能源效率。 - Developers connect through several APIs.
開發者透過多個 API 進行連接。 - DSX Sim.
DSX Sim。 - For physical.
用於物理。 - Electrical.
電力。 - Thermal.
散熱。 - And network simulation.
以及網路模擬。 - DSX Exchange.
DSX Exchange。 - For AI factory operational data.
用於 AI 工廠營運資料。 - DSX Flex.
DSX Flex。 - For secure dynamic power management between the grid.
用於電網之間的安全動態電力管理。 - And DSX Max-Q.
以及 DSX Max-Q。 - To dynamically maximize token throughput.
用於動態最大化 token 吞吐量。
AI 工廠設計與營運流程
- It starts with Sim ready assets from NVIDIA and equipment manufacturers.
從 NVIDIA 和設備製造商的模擬就緒資產開始。 - Managed by PTC Windchill PLM.
由 PTC Windchill PLM 管理。 - Then model based systems engineering is done in Daciland. 3D experience.
然後在 Dassault 的 3DEXPERIENCE 中進行基於模型的系統工程。 - Jacobs brings the data into their custom omniverse app to finalize design.
Jacobs 將資料匯入他們自訂的 Omniverse 應用程式以完成設計。 - It’s tested with leading simulation tools.
透過領先的模擬工具進行測試。 - Using Siemens Star CCM Plus for external thermals.
使用 Siemens Star-CCM+ 進行外部散熱模擬。 - Cadence Reality for internal.
Cadence Reality 用於內部散熱。 - E-Tap for electrical.
E-Tap 用於電力模擬。 - And Nvidia’s network simulator DSX Air.
以及 NVIDIA 的網路模擬器 DSX Air。 - And virtually commissioned through Procore to ensure accelerated construction time.
並透過 Procore 進行虛擬驗收，以確保加速施工時間。 - When the site goes live, the digital twin becomes the operator.
當站點上線時，數位孿生成為營運者。 - AI agents work with DSX Max-Q to dynamically orchestrate infrastructure.
AI 代理與 DSX Max-Q 協同工作，動態調度基礎設施。 - Phaedra’s agent oversees cooling and electrical systems.
Phaedra 的代理負責監控散熱和電力系統。 - Sending signals to Max-Q which continuously optimizes compute throughput and energy efficiency.
向 Max-Q 發送訊號，持續最佳化運算吞吐量和能源效率。 - Emerald AI agents interpret live grid demand and stress signals.
Emerald AI 代理解讀即時電網需求和壓力訊號。 - And adjust power dynamically.
並動態調整電力。
生態系與基礎設施目標
- With DSX, Nvidia and our ecosystem of partners are racing to build AI infrastructure around the world.
透過 DSX，NVIDIA 和我們的生態系合作夥伴正在競相於全球建造 AI 基礎設施。 - Ensuring extreme resiliency, efficiency, and throughput.
確保極致的韌性、效率和吞吐量。 - It’s incredible, right?
非常驚人，對吧？
Omniverse 與 AI 工廠平台
- Well, Omniverse was designed to hold the world’s digital twin.
Omniverse 的設計初衷是承載全世界的數位孿生。 - Starting from the Earth.
從地球開始。 - And it’s going to hold digital twins of all sizes.
它將承載各種規模的數位孿生。 - And so we have just such a great ecosystem of partners.
因此我們擁有如此優秀的合作夥伴生態系。 - I want to thank all of you.
我想感謝你們所有人。 - All of these companies are brand new to our world.
所有這些公司對我們的世界來說都是全新的。 - We didn’t know many of you just a couple years ago.
就在幾年前我們還不認識你們當中的許多人。 - And now we’re working so close together to work on and build together the largest computer.
而現在我們緊密合作，共同打造有史以來最大的電腦。 - The world’s ever seen.
世界上前所未見的。 - And also to do it at planetary scale.
而且是在行星規模上實現。 - So Nvidia DSX is our new AI factory platform.
所以 NVIDIA DSX 是我們全新的 AI 工廠平台。 - I’ll spend very little time on this this time.
這次我不會花太多時間在這上面。
太空運算與資料中心
- However, we’re going to space.
然而，我們即將前往太空。 - We’ve already been out in space.
我們已經在太空中了。 - Thor is radiation approved.
Thor 已獲得抗輻射認證。 - And we’re in satellites.
我們已在衛星中部署。 - You do imaging from satellites in the future.
未來你可以從衛星進行影像處理。 - We’ll also build data centers in space.
我們還將在太空中建造資料中心。 - Obviously very complicated to do so.
顯然這非常複雜。 - We’re working with our partners on a new computer called Vera Rubin Space 1.
我們正與合作夥伴共同開發一款名為 Vera Rubin Space 1 的新電腦。 - And it’s going to go out to space and start data centers out in space.
它將前往太空，並在太空中啟動資料中心。 - Now, of course, in space there’s no conduction.
當然，在太空中沒有傳導散熱。 - There’s no convection.
沒有對流散熱。 - There’s just radiation.
只有輻射散熱。 - And so we have to figure out how to cool these systems out in space.
因此我們必須想出如何在太空中為這些系統散熱。 - But we’ve got lots of great engineers working on it.
但我們有許多優秀的工程師正在努力解決這個問題。 - Let me tell you a little bit about the future of data.
讓我來談談資料的未來。
5️⃣ OpenClaw Agent 革命
支持 OpenClaw 專案
- So Peter Steinberger is here.
Peter Steinberger 在這裡。 - And he wrote a piece of software.
他寫了一個軟體。 - It’s called OpenClaw.
它叫做 OpenClaw。 - And I don’t know if he realized how successful it’s going to be.
我不知道他是否意識到它會有多成功。 - But the importance of data is that it’s going to be able to do things.
但資料的重要性在於它能夠做到很多事。 - And the importance is profound.
這個重要性是深遠的。 - OpenClaw is the number one, it’s the most popular open source project in the history of humanity.
OpenClaw 是第一名，它是人類歷史上最受歡迎的開源專案。 - And it did so in just a few weeks.
而且它只花了幾週就做到了。 - It exceeded what Linux did in 30 years.
它超越了 Linux 30 年來的成就。 - And it’s that important.
它就是這麼重要。 - It is that important.
它確實就是這麼重要。 - It will do well.
它會發展得很好。 - This is all you do.
你只需要做這些。 - Okay?
好嗎？ - So I’m announcing our support of it.
所以我宣布我們對它的支持。
OpenClaw 功能與應用
- Let me just quickly go through this.
讓我快速瀏覽一下。 - I want to show you a couple of things.
我想展示幾件事情。 - You simply type this.
你只需要輸入這個。 - You type this into a console.
你在主控台輸入這個。 - And it goes out.
然後它就開始執行。 - It finds OpenClaw.
它會找到 OpenClaw。 - It downloads it.
它會下載它。 - It builds you an AI agent.
它為你建立一個 AI agent。 - And then you can tell it whatever else you need to do.
然後你可以告訴它你還需要做什麼。 - Okay?
好嗎？ - So let’s take a look.
讓我們來看看。 - Research is a huge deal.
研究是非常重要的事。 - You give an AI agent a task, go to sleep.
你給 AI agent 一個任務，然後去睡覺。 - It runs 100 experiments overnight, keeping what works and killing what doesn’t.
它在夜間執行 100 個實驗，保留有效的，淘汰無效的。 - I really love what my stuff enables that person to do.
我真的很喜歡我的東西讓那個人能做到的事。 - And I had like one guy, he told me like he installed it as a 60-year-old dad and like they made beer, connected the machine via Bluetooth.
有一個人告訴我，他是個 60 歲的爸爸，安裝了它，然後他們釀啤酒，透過藍牙連接機器。 - And then we automated everything, including the whole website for people to order.
然後我們自動化了一切，包括整個讓人們下單的網站。 - The lobster lager.
龍蝦拉格啤酒。 - Hundreds of people are queuing up for lobsters in St. Jeff.
數百人在 St. Jeff 排隊等著買龍蝦。 - OpenClaw.
OpenClaw。 - OpenClaw.
OpenClaw。 - We want to build OpenClaw with OpenClaw.
我們想用 OpenClaw 來建造 OpenClaw。 - Everyone is talking about OpenClaw, but what the f* is OpenClaw?
每個人都在談論 OpenClaw，但 OpenClaw 到底是什麼？ - Believe it or not, there’s already a claw con.
信不信由你，已經有一個 CLAW Con 了。 - OpenClaw.
OpenClaw。 - OpenClaw.
OpenClaw。 - OpenClaw.
OpenClaw。 - OpenClaw.
OpenClaw。 - OpenClaw.
OpenClaw。 - Incredible.
不可思議。 - Incredible.
不可思議。 - Now, I illustrated effectively what OpenClaw is in this way, so that all of you can understand it.
現在，我用這種方式有效地說明了 OpenClaw 是什麼，讓你們所有人都能理解。
OpenClaw 作為自主式作業系統
- But let’s just think what happened.
但讓我們想想發生了什麼。 - What is OpenClaw?
OpenClaw 是什麼？ - It connects.
它能連接。 - It’s an agentic system.
它是一個自主式系統。 - It calls and connects to large language models.
它呼叫並連接到大型語言模型。 - So, the first thing it has, it has resources.
首先，它擁有資源。 - It has the resources that it manages.
它有它所管理的資源。 - It could access tools.
它能存取工具。 - It could access file systems.
它能存取檔案系統。 - It could access large language models.
它能存取大型語言模型。 - It’s able to do scheduling.
它能做排程。 - It’s able to do cron jobs.
它能做定時任務。 - It’s able to decompose a problem, a prompt that you gave it, into step by step by step.
它能把一個問題、你給它的一個提示，拆解成一步一步一步。 - It could spawn off and call upon other sub-agents.
它能產生並呼叫其他子代理。 - It has I.O.
它有 I/O。 - You could talk to it in any modality you want.
你可以用任何你想要的模態與它對話。 - You could wave at it and it understands you.
你可以對它揮手，它就能理解你。 - You could talk to any modality you want.
你可以用任何你想要的模態對話。 - It sends you messages.
它會發訊息給你。 - It texts you, sends you email.
它傳簡訊給你、寄電子郵件給你。 - So, it’s got I.O.
所以，它有 I/O。 - What else does it have?
它還有什麼？ - Well, based on that, you could say, in fact, it’s an operating system.
基於以上這些，你可以說，事實上，它是一個作業系統。 - I’ve just used the same syntax that I would describe an operating system.
我剛剛用了描述作業系統的相同語法。 - OpenClaw has open sourced, essentially, the operating system of agentic computers.
OpenClaw 基本上開源了自主式電腦的作業系統。 - It is no different than how Windows made it possible for us to create personal computers.
這和 Windows 讓我們得以打造個人電腦沒有什麼不同。 - Now, OpenClaw has made it possible for us to create personal agents.
現在，OpenClaw 讓我們得以打造個人 agent。 - The implication is incredible.
其影響是不可思議的。 - The implication is incredible.
其影響是不可思議的。
OpenClaw 的戰略重要性
- First of all, the adoption says something, you know, all in itself.
首先，採用率本身就說明了一些事情。 - However, the most important thing is this.
然而，最重要的事情是這個。 - Every single company now realizes, every single company, every single software company, every single technology company, for the CEOs, the question is, what’s your OpenClaw strategy?
每一家公司現在都意識到，每一家公司、每一家軟體公司、每一家科技公司，對 CEO 來說，問題是：你的 OpenClaw 策略是什麼？ - Just as we need to all have a Linux strategy, we all need to have an HTTP, HTML strategy, which started the internet.
就像我們都需要有 Linux 策略一樣，我們都需要有 HTTP、HTML 策略，那是開啟網際網路的關鍵。 - We all need to have a Kubernetes strategy, which made it possible for mobile cloud to happen.
我們都需要有 Kubernetes 策略，那是讓行動雲端成為可能的關鍵。 - Every company in the world today needs to have an OpenClaw strategy, an agentic system strategy.
今天世界上每一家公司都需要有 OpenClaw 策略，一個自主式系統策略。 - This is the number one thing that we need to do.
這是我們需要做的第一件事。 - This is the new computer.
這就是新的電腦。
企業 IT 轉型
自主式系統推動企業 IT 演進
- Now, this is just the exciting part.
現在，這只是令人興奮的部分。 - This is enterprise I.T. before OpenClaw.
這是 OpenClaw 之前的企業 IT。 - You know, and I mentioned earlier, the way enterprise I.T. works.
你知道的，我之前提到過企業 IT 的運作方式。 - And the reason why it’s called data centers is because these large rooms, these large buildings held data, held the files of people, the structured data of business.
它之所以被稱為資料中心，是因為這些大型房間、這些大型建築物儲存著資料、人們的檔案、企業的結構化資料。 - It would pass through software that has tools and, you know, systems of records and all kinds of workflow that’s codified into it.
它會經過擁有工具的軟體，你知道的，記錄系統和各種被編碼進去的工作流程。 - And that turns into tools that humans would use, digital workers would use.
然後變成人類會使用的工具、數位工作者會使用的工具。 - That is the old I.T. industry.
那是舊的 IT 產業。 - Software companies creating tools, saving files, and, of course, G.S.I.s consultants that help companies figure out how to use these tools and integrate these tools.
軟體公司建立工具、儲存檔案，當然還有 GSI 顧問幫助企業弄清楚如何使用這些工具和整合這些工具。 - These tools are incredibly valuable for governance and security and privacy and compliance, and all of that continues to be true.
這些工具對治理、安全、隱私和合規非常有價值，這一切仍然是事實。 - It’s just that post OpenClaw, post agentic, this is what it’s going to look like.
只是在 OpenClaw 之後、在自主式之後，它將會變成這個樣子。 - This is the extraordinary part.
這是非凡的部分。 - Every single I.T. company, every single company, every S.A.A.S. company, every S.A.A.S. company will become a G.A.A.S. company.
每一家 IT 公司、每一家公司、每一家 SaaS 公司，每一家 SaaS 公司都將成為 GaaS 公司。 - No question about it.
毫無疑問。 - Every single S.A.A.S. company will become a G.A.A.S. company, an agentic as a service company.
每一家 SaaS 公司都將成為 GaaS 公司，一個自主式即服務的公司。 - And what’s amazing is this, you now, OpenClaw gave us, gave the industry exactly what it needed at exactly the time.
令人驚嘆的是，OpenClaw 在恰當的時機給了我們、給了整個產業恰好需要的東西。 - Just as Linux gave the industry exactly what it needed at exactly the time, just as Kubernetes showed up at exactly the right time, just as H.T.M.L. showed up.
就像 Linux 在恰當的時機給了產業所需的一切，就像 Kubernetes 在恰當的時機出現，就像 HTML 出現一樣。 - It made it possible for the entire industry to grab onto this open source stack and go do something with it.
它讓整個產業得以抓住這個開源技術堆疊，用它去做些什麼。
自主式系統安全考量
- There’s just one catch.
只有一個問題。 - Agentic systems in the corporate network can have access to sensitive information, can have access to sensitive information, it can execute code, and it can communicate externally.
企業網路中的自主式系統可以存取敏感資訊、可以存取敏感資訊、可以執行程式碼，而且可以對外通訊。
安全與隱私顧慮
- Just say that out loud.
大聲說出來。 - Okay, think about it.
好，想想看。 - Access sensitive information, execute code, communicate externally.
存取敏感資訊、執行程式碼、對外通訊。 - You could of course access employee information, access supply chain, access finance information, sensitive information, and send it out.
你當然可以存取員工資訊、存取供應鏈、存取財務資訊、敏感資訊，然後把它發送出去。 - Communicate externally.
對外通訊。 - Communicate externally.
對外通訊。 - Obviously, obviously, this can’t possibly be allowed.
很明顯，很明顯，這絕不能被允許。
企業安全與隱私的 OpenClaw
- And so, what we did was, we worked with Peter.
所以，我們所做的是，我們與 Peter 合作。
NVIDIA OpenClaw 參考設計
- We took some of the world’s best security and computing experts, and we worked with Peter to make OpenClaw, OpenClaw, enterprise secure and enterprise private capable.
我們集結了一些世界上最頂尖的安全和運算專家，與 Peter 合作讓 OpenClaw 具備企業級安全和企業級隱私能力。 - And we call that, this is our NVIDIA OpenClaw reference for Open NemoClaw, which is a reference for OpenClaw, and it has all these agentic AI toolkits.
我們稱之為 NVIDIA OpenClaw 參考設計，用於 Open NemoClaw，它是 OpenClaw 的參考設計，包含所有這些自主式 AI 工具套件。 - And the first part of it is technology we call Open Shell, that has now been integrated into OpenClaw.
其中第一個部分是我們稱為 Open Shell 的技術，它現在已經被整合到 OpenClaw 中。 - Now it’s enterprise ready.
現在它已經準備好應用於企業了。
策略引擎與網路防護
- This stack, this stack, with a reference design we call NemoClaw, okay, with a reference stack we call NemoClaw.
這個技術堆疊，這個技術堆疊，搭配我們稱為 NemoClaw 的參考設計，好的，搭配我們稱為 NemoClaw 的參考堆疊。 - You could download it, play with it, and you could connect to it the policy engine of all of the SAS companies in the world.
你可以下載它、使用它，而且你可以把世界上所有 SaaS 公司的策略引擎連接到它上面。 - And your policy engines are super important, super valuable.
你的策略引擎非常重要、非常有價值。 - So the policy engines could be connected, NemoClaw, or OpenClaw with Open Shell would be able to execute that policy engine.
所以策略引擎可以被連接，NemoClaw 或是搭配 Open Shell 的 OpenClaw 將能夠執行該策略引擎。 - It has a, it has a network guardrail.
它有網路護欄。 - It has a privacy router.
它有隱私路由器。 - And as a result, we could protect and keep the, the CLAWs from executing inside our company and do it safely.
因此，我們可以保護並確保 CLAW 在公司內部安全地執行。
NVIDIA 開放模型倡議
- We also added several things to the agentic system.
我們也在自主式系統中加入了幾項功能。 - And one of the most important things you want to do with your own CLAW, custom CLAWs, is so that you can have your custom models.
你在自己的 CLAW、客製化 CLAW 上最想做的重要事情之一，就是擁有你自己的客製化模型。 - And this is NVIDIA’s open model initiative.
這就是 NVIDIA 的開放模型倡議。 - We are now at the frontier of every single domain of AI models.
我們現在處於每一個 AI 模型領域的前沿。
多元 AI 模型領域
- Whether it’s Nemotron, Cosmos World Foundation model, Groot, artificial general robotics, human robotics models, Alpamayo for autonomous vehicle, BioNemo for digital biology, ERT2 for AI physics.
無論是 Nemotron、Cosmos 世界基礎模型、Groot、通用機器人、人形機器人模型、用於自動駕駛的 Alpamayo、用於數位生物學的 BioNemo、用於 AI 物理的 Earth2。 - We are at the frontier on every single one.
我們在每一個領域都處於前沿。 - Take a look.
看一看。 - The world is diverse.
世界是多元的。 - No single model can serve every industry.
沒有單一模型可以服務每個產業。
關鍵開放模型系列
- Open models, is one of the largest and most diverse AI ecosystems in the world.
開放模型，是世界上最大且最多元的 AI 生態系統之一。 - Nearly 3 million open models across language, vision, biology, physics, and autonomous systems enable AI builds for specialized domains.
近 300 萬個橫跨語言、視覺、生物、物理和自主系統的開放模型，讓特定領域的 AI 建構成為可能。 - NVIDIA is one of the largest contributors to open source AI.
NVIDIA 是開源 AI 最大的貢獻者之一。 - We build and release six families of open frontier models, plus the training data, recipes, and frameworks to help developers customize and adopt.
我們建構並發布六個系列的開放前沿模型，加上訓練資料、配方和框架，幫助開發者客製化和採用。 - New leaderboard types, new leaderboard topping models are launching for every family.
新的排行榜類型、新的排行榜冠軍模型正在每個系列中推出。 - At the core, Nemotron, reasoning models for language, visual understanding, RAG, safety, and speech.
核心是 Nemotron，用於語言、視覺理解、RAG、安全和語音的推理模型。
開放模型與 Nemotron
開放模型的進展與承諾
- Can you hear me now?
你現在能聽到我嗎？ - Yes.
是的。 - Hello?
哈囉？ - Yes, I can hear you now.
是的，我現在能聽到你了。 - Cosmos, frontier models for physical AI world generation and understanding.
Cosmos，用於物理 AI 世界生成與理解的前沿模型。 - Alpamayo, the world’s first thinking and reasoning autonomous vehicle AI.
Alpamayo，世界上第一個具有思考和推理能力的自動駕駛 AI。 - Groot, foundation models for general purpose robots.
Groot，用於通用機器人的基礎模型。 - BioNemo, open models for biology, chemistry, and molecular design.
BioNemo，用於生物學、化學和分子設計的開放模型。 - Earth2, models for weather and climate forecasting rooted in AI physics.
Earth2，植根於 AI 物理的天氣和氣候預測模型。 - NVIDIA open models give researchers and developers the foundation to build and build AI systems.
NVIDIA 開放模型為研究人員和開發者提供了建構 AI 系統的基礎。 - We’re building the world’s first thinking and reasoning foundation to build and deploy AI for their own specialized domains.
我們正在建構世界上第一個思考和推理基礎，讓人們能為自己的專業領域建構和部署 AI。 - Our models, our mo- Thank you.
我們的模型，我們的模… 謝謝。 - Our models are valuable to all of you because number one, it’s on the top of the leaderboard.
我們的模型對你們所有人都有價值，因為第一，它在排行榜的頂端。 - It’s world class.
它是世界級的。 - But most importantly, it’s because we are not going to give up working on it.
但最重要的是，因為我們不會放棄持續開發它。
持續模型演進
- We’re going to keep on working on it every single day.
我們每一天都會持續開發它。 - Nemotron 3 is going to be followed by Nemotron 4.
Nemotron 3 之後會有 Nemotron 4。 - Cosmos 1 is followed by Cosmos 2.
Cosmos 1 之後是 Cosmos 2。 - Groot, Groot at generation 2.
Groot，Groot 第 2 代。 - Each and one of these will continue to advance these models.
這些模型的每一個都將持續進步。
垂直整合與水平開放
- Vertical integration, horizontal openness, so that we can enable everybody to join the AI revolution.
垂直整合、水平開放，讓我們能夠使每個人都加入 AI 革命。 - Number one on leaderboard across research and voice and world models and artificial general robotics and self-driving cars and reasoning.
在研究、語音、世界模型、通用機器人、自動駕駛和推理等領域的排行榜上都是第一名。 - And of course, one of the most important ones.
當然，其中最重要的之一。 - This is Nemotron 3 in OpenClaw.
這是 OpenClaw 中的 Nemotron 3。 - This is Nemotron 3 in OpenClaw.
這是 OpenClaw 中的 Nemotron 3。 - And look at the top three.
看看前三名。 - There are the three best models in the world.
那是世界上最好的三個模型。 - Okay?
好嗎？ - So, we are at the frontier.
所以，我們在前沿。
客製化與主權 AI
- It is also true, it is also true that we want to create the foundation model so that all of you can fine tune it, post-train it into exactly the intelligence you need.
同樣確實的是，我們想要建立基礎模型，讓你們所有人都能微調它、後訓練它，使其成為你們所需要的精確智慧。 - This is Nemotron 3 Ultra.
這是 Nemotron 3 Ultra。 - It is going to be the best base model the world has ever created.
它將是世界上有史以來最好的基礎模型。 - This allows us to help every country build their sovereign AI.
這讓我們能幫助每個國家建構自己的主權 AI。
Nemotron 聯盟與合作夥伴
- And we are working with so many different companies out there.
我們正在與許多不同的公司合作。 - And one of the most exciting things that we are doing today, I am announcing today, is a Nemotron coalition.
今天我們正在做的最令人興奮的事情之一，我今天宣布的是 Nemotron 聯盟。 - We are so dedicated to this.
我們對此全力以赴。 - We have invested billions of dollars of AI infrastructure so that we could develop the core engines for AI that is necessary for all the libraries of inference and so on.
我們投資了數十億美元的 AI 基礎設施，以便我們能開發所有推理程式庫等所需的 AI 核心引擎。 - But also to create the AI models to activate every single industry in the world.
同時也是為了建立 AI 模型來啟動世界上每一個產業。 - Large language models is really important.
大型語言模型真的很重要。 - Of course it is important.
當然它很重要。 - How could human intelligence not be?
人類智慧怎麼可能不重要？ - However, in different industries around the world, in different countries around the world, you need to have the ability to customize your own models and the domains that are there.
然而，在世界各地不同的產業中、在世界各地不同的國家中，你需要有能力客製化自己的模型和相關的領域。 - The domain of the models is radically different.
模型的領域截然不同。 - From biology, to physics, to self-driving cars, to general robotics, to of course human language.
從生物學、到物理學、到自動駕駛、到通用機器人，當然還有人類語言。 - And we have the ability to work with every single region to create their domain specific, their sovereign AI.
我們有能力與每一個地區合作，建立他們的領域專用、他們的主權 AI。 - Today, we are announcing a coalition to partner with us to make Nemotron 4 even more amazing.
今天，我們宣布一個聯盟，與我們合作讓 Nemotron 4 更加出色。 - And that coalition has some amazing companies in it.
這個聯盟中有一些很棒的公司。 - Black Forest Labs, Imaging Company, Cursor, the famous coding company, we use lots of it.
Black Forest Labs 影像公司、Cursor 知名的程式撰寫公司，我們大量使用它。 - Langchain, billion downloads for creating custom agents.
Langchain，用於建立客製化 agent 的十億次下載量。 - Mistral, Arthur mentioned, I think he’s here.
Mistral，Arthur 提到過，我想他在這裡。
領導企業聯盟
- Incredible, incredible company.
不可思議的公司。 - Perplexity, Perplexi’s computer.
Perplexity，Perplexity 的電腦。 - Absolutely use it.
一定要用它。 - Everybody use it.
每個人都用它。 - It is so good.
它非常好。 - A multi-modal agentic system.
一個多模態自主式系統。 - Reflection, Sarvam from India, Thinking Machine, Miramarati’s lab.
Reflection、來自印度的 Sarvam、Thinking Machine、Miramarati 的實驗室。 - Incredible companies.
不可思議的公司。 - Incredible companies joining us.
不可思議的公司加入我們。 - Thank you.
謝謝。
企業自主式轉型
企業採用自主式系統
- I said, I said that every single enterprise company, every single software company in the world needs an agentic systems, need an agent strategy.
我說過，我說過世界上每一家企業公司、每一家軟體公司都需要自主式系統，需要 agent 策略。 - You need to have an open-claw strategy.
你需要有 OpenClaw 策略。 - And they all agree.
他們都同意。 - And they’re all partnering with us to integrate Nemo, the NemoClaw reference design, the NVIDIA agentic AI toolkit, and of course, all of our open models.
他們都與我們合作整合 Nemo、NemoClaw 參考設計、NVIDIA 自主式 AI 工具套件，當然還有我們所有的開放模型。 - One company after another.
一家接一家。 - There’s so many.
有很多。 - And we’re partnering with all of you.
我們與你們所有人合作。 - I’m really grateful for that.
我對此非常感激。
企業 IT 的文藝復興
- And this is our moment.
這是我們的時刻。 - This is a reinvention.
這是一次重塑。 - This is a renaissance.
這是一次文藝復興。 - A renaissance of the enterprise IT.
企業 IT 的文藝復興。 - From what would be a $2 trillion industry, this is going to become a multi-trillion dollar industry.
從一個 2 兆美元的產業，這將成為一個數兆美元的產業。 - Offering not just tools for people to use, but agents that are specialized in very special domains that you’re expert in, that we could rent.
提供的不僅僅是供人使用的工具，還有專精於你所擅長的特定領域的 agent，我們可以租用。
工程師生產力與 Token 預算
- I could totally imagine in the future, every single engineer in our company will need an annual token budget.
我完全可以想像在未來，我們公司的每一位工程師都需要一個年度 token 預算。 - They’re going to make a few hundred thousand dollars a year, their base pay.
他們一年會賺幾十萬美元的底薪。 - I’m going to give them probably half of that on top of it, as tokens, so that they could build a company that’s going to be able to do that.
我可能會在此之上再給他們大約一半的金額作為 token，讓他們能夠建構一個有能力做到這些事的公司。 - So that they could be Amplify 10X.
讓他們能夠放大 10 倍。
Token 作為招募與生產力工具
- Of course we would.
我們當然會。 - It is now one of the recruiting tools in Silicon Valley.
它現在是矽谷的招募工具之一。 - How many tokens comes along with my job?
我的工作附帶多少 token？ - And the reason for that is very clear.
原因非常清楚。 - Because every engineer that has access to tokens will be more productive.
因為每一位能使用 token 的工程師都會更有生產力。 - And those tokens, as you know, will be produced by AI factories that all of you and us, we partner to build.
而那些 token，如你們所知，將由我們所有人與你們合作建構的 AI 工廠生產。
企業與軟體公司的未來
- So every single enterprise company in today sit on top of file systems and data centers.
所以今天每一家企業公司都建立在檔案系統和資料中心之上。 - Every single software company of the future will be agentic, and they will be token manufacturers.
未來每一家軟體公司都將是自主式的，它們將成為 token 製造商。 - They’ll be token users for their engineers, and they’ll be token manufacturers for all of their customers.
它們將為工程師提供 token 使用，也將為所有客戶製造 token。 - The open clause event, the open clause event cannot be understated.
OpenClaw 事件，OpenClaw 事件的重要性不可低估。
OpenClaw 無可比擬的重要性
- This is as big of a deal as HTML.
這與 HTML 一樣重大。 - This is as big of a deal as Linux.
這與 Linux 一樣重大。 - We have now a world-class open agentic framework that all of us could use to build our open clause strategy.
我們現在擁有一個世界級的開放自主式框架，我們所有人都能用它來建構我們的 OpenClaw 策略。 - And we’ve created a reference design we call NemoClaw, that all of you could use, that is optimized, it’s performant, it is safe and secure.
我們建立了一個稱為 NemoClaw 的參考設計，你們所有人都能使用，它是最佳化的、高效能的、安全且可靠的。
6️⃣ 物理 AI
理解 Agent：數位 vs. 物理
- Speaking of agents, agents as you know, perceive, reason, and act.
說到 agent，如你們所知，agent 能感知、推理和行動。 - Most of the agents in the world today that I’ve spoken about are digital agents.
我今天談到的世界上大多數 agent 都是數位 agent。 - They act in the digital world.
它們在數位世界中行動。 - They reason, they write software.
它們推理、撰寫軟體。 - It’s all digital.
一切都是數位的。 - But we also have been working on physically embodied agents for a long time.
但我們也已經在物理實體 agent 上耕耘了很長時間。
物理實體 Agent：機器人
- We call them robots.
我們稱它們為機器人。 - And the AIs that they need are physical AIs.
而它們所需要的 AI 就是物理 AI。 - We have some big announcements here.
我們在這裡有一些重大發表。
機器人生態系與合作夥伴
- I’m going to just walk through a few of them. 110 robots here.
我就簡單帶過其中幾個。這裡有 110 台機器人。 - Almost every single company in the world, I can’t think of one, that are building robots is working with NVIDIA.
全世界幾乎每一家在做機器人的公司，我想不到例外，都在與 NVIDIA 合作。
機器人運算基礎設施
- We have three computers.
我們有三台電腦。 - The training computer, the synthetic data generation and simulation computer, and of course the robotics computer that sits inside the robot itself.
訓練電腦、合成資料生成與模擬電腦，當然還有裝在機器人內部的機器人電腦。 - We have all the software stacks necessary to do so.
我們擁有所有必要的軟體堆疊來實現這一切。
軟體與生態系整合
- The AI models to help you.
能幫助你的 AI 模型。 - And all of this is integrated into ecosystems around the world.
而這一切都整合進全球各地的生態系中。 - And all of our partners from Siemens to Cadence.
以及我們所有的合作夥伴，從 Siemens 到 Cadence。
新機器人合作夥伴
- Incredible partners everywhere.
到處都是了不起的合作夥伴。 - And today we’re announcing a whole bunch of new partners.
今天我們要宣布一大批新合作夥伴。
自動駕駛車輛與 RoboTaxi 部署
- As you know, we’ve been working on self-driving cars for a long time.
如你們所知，我們在自動駕駛車輛上已經耕耘了很長時間。 - The chat GPT moment of self-driving cars has arrived.
自動駕駛車輛的 ChatGPT 時刻已經到來。 - We now know we could successfully, autonomously drive cars.
我們現在知道我們能成功地自主駕駛汽車。 - And today we are announcing four new partners for NVIDIA’s RoboTaxi Ready platform.
今天我們宣布 NVIDIA RoboTaxi Ready 平台的四個新合作夥伴。 - B-Y-D, Hyundai, Nissan, Geely.
BYD、Hyundai、Nissan、Geely。 - All together, 18 million cars built each year.
加在一起，每年生產 1,800 萬輛車。 - Joining our partners from before, Mercedes, Toyota, GM.
加入我們先前的合作夥伴 Mercedes、Toyota、GM。 - The number of RoboTaxi Ready cars in the future are going to be incredible.
未來 RoboTaxi Ready 車輛的數量將會非常驚人。 - And we’re announcing also a big partnership with Uber.
我們也宣布與 Uber 的重大合作夥伴關係。 - Multiple cities, we’re going to be deploying and connecting these RoboTaxi Ready vehicles into their network.
我們將在多個城市部署這些 RoboTaxi Ready 車輛，並將它們連接到 Uber 的網路中。 - And so, a whole bunch of new cars.
所以，一大批新車。
更多機器人公司合作夥伴
- We have ABB, Universal Robotics, KUKA.
我們有 ABB、Universal Robotics、KUKA。 - So many robotics companies here.
這裡有這麼多機器人公司。
物理 AI 整合至製造與基礎設施
- And we’re working with them to implement our physical AI models integrated into simulation systems so that we could deploy these robots into manufacturing lines all over.
我們正與他們合作，將我們的物理 AI 模型整合進模擬系統中，以便將這些機器人部署到各地的生產線上。 - We have Caterpillar here.
我們這裡有 Caterpillar。 - We even have T-Mobile here.
我們這裡甚至有 T-Mobile。 - And the reason for that is in the future, that radio tower used to be a radio tower, is going to be an NVIDIA Aerial AI RAM.
原因是在未來，那座過去只是無線電塔的東西，將會變成 NVIDIA Aerial AI RAM。 - And so, this is going to be a robotics radio tower.
所以，這將會是一座機器人無線電塔。 - Meaning, it can reason about the traffic, figure out how to adjust its beam forming so that it could save as much energy as possible and increase the amount of fidelity as possible.
意思是，它能推理流量狀況，找出如何調整波束成形，以盡可能節省能源並盡可能提高訊號品質。 - There are so many humanoid robots here.
這裡有這麼多人形機器人。 - But one of my favorites, one of my favorites, is a Disney robot.
但我最喜歡的其中一個，我最喜歡的其中一個，是 Disney 機器人。 - You know what?
你們知道嗎？
物理 AI 與機器人展示
- Tell you what.
跟你們說。 - Let me just show you some of the videos.
讓我放一些影片給你們看。 - Let’s look at that first.
我們先看這個。
自動駕駛車輛
- The first global rollout of physical AI at scale is here.
物理 AI 首次大規模全球推出就在這裡。
物理 AI 與機器人時代
- Autonomous vehicles.
自動駕駛車輛。 - And with NVIDIA Alpamayo, vehicles now have reasoning, helping them operate safely and intelligently across scenarios.
透過 NVIDIA Alpamayo，車輛現在具備推理能力，幫助它們在各種場景中安全且智慧地運作。 - We ask the car to narrate its actions.
我們要求車輛描述自己的行動。 - I’m changing lanes to the right to follow my route.
我正在向右變換車道以跟隨我的路線。 - Explain its thinking as it makes decisions.
在做決策時解釋它的思考過程。 - There’s a double parked vehicle in my lane.
我的車道上有一輛違規停放的車。 - I’m going around it.
我正在繞過它。 - And follow instructions.
並且遵循指令。 - Hey Mercedes, can you speed up?
嘿 Mercedes，你能加速嗎？ - Sure, I’ll speed up.
好的，我會加速。
機器人訓練與模擬
- This is the age of physical AI and robotics.
這是物理 AI 與機器人的時代。
機器人訓練：從模擬到現實
真實世界機器人訓練的挑戰
- Around the world, developers are building robots of every kind.
全球各地的開發者正在建造各種機器人。 - But the real world is massively diverse.
但真實世界極度多樣。 - Unpredictable.
不可預測。 - Full of edge cases.
充滿邊界案例。 - Real world data will never be enough to train for every scenario.
真實世界的資料永遠不足以訓練所有場景。 - We need data generated from AI and simulation.
我們需要由 AI 和模擬生成的資料。
運算即資料與模擬
- For robots, compute is data.
對機器人而言，運算就是資料。 - Developers pre-train World Foundation models on internet-scale video and human demonstrations and evaluate the model’s performance to prepare them for post-training.
開發者在網路規模的影片和人類示範上預訓練 World Foundation 模型，並評估模型的表現以準備後續訓練。 - Using classical and neural simulation, they generate massive amounts of synthetic data and train policies at scale.
透過經典和神經模擬，他們生成大量的合成資料並大規模訓練策略。
NVIDIA 機器人訓練開源工具
- To accelerate developers, NVIDIA built open-source Isaac Lab for robot training and evaluation and simulation.
為了加速開發者的進展，NVIDIA 建立了開源的 Isaac Lab，用於機器人訓練、評估與模擬。 - Newton for extensible and GPU-accelerated differentiable physics simulation.
Newton 用於可擴展且 GPU 加速的可微分物理模擬。 - Cosmos World Models for neural simulation.
Cosmos World Models 用於神經模擬。 - And Groot Open Robotics Foundation models for robot reasoning and action generation.
以及 Groot Open Robotics Foundation 模型，用於機器人推理和動作生成。 - With enough compute, developers everywhere are closing the physical AI data gap.
有了足夠的運算資源，各地的開發者正在縮小物理 AI 的資料差距。
開發者成功案例
- Peritas AI trains their own robots.
Peritas AI 訓練他們自己的機器人。 - NVIDIA Isaac Lab trains their operating room assistant robot in NVIDIA Isaac Lab, multiplying their data with NVIDIA Cosmos World Models.
他們在 NVIDIA Isaac Lab 中訓練手術室助理機器人，透過 NVIDIA Cosmos World Models 倍增資料。 - Skilled AI uses Isaac Lab and Cosmos to generate post-training data for their skilled AI brain.
Skilled AI 使用 Isaac Lab 和 Cosmos 為其 Skilled AI 大腦生成後續訓練資料。 - They use reinforcement learning to harden the model across thousands of variations.
他們使用強化學習在數千種變化中強化模型。 - Humanoid uses Isaac Lab to train whole body control and manipulation policies.
Humanoid 使用 Isaac Lab 來訓練全身控制和操作策略。 - Hexagon Robotics uses Isaac Lab for training and data generation.
Hexagon Robotics 使用 Isaac Lab 進行訓練和資料生成。 - Foxconn fine-tunes group models in Isaac Lab, as does Noble Machines.
Foxconn 在 Isaac Lab 中微調 Groot 模型，Noble Machines 也是如此。 - Disney Research uses their Camino physics simulator in Newton and Isaac Lab to train policies across their character robots in every universe.
Disney Research 使用他們的 Camino 物理模擬器搭配 Newton 和 Isaac Lab，在每個宇宙中訓練其角色機器人的策略。
Disney 機器人展示
- Ladies and gentlemen, Olaf.
女士們先生們，Olaf。 - Wahoo!
哇呼！ - Snowman coming through.
雪人來了。 - Newton works.
Newton 有效。 - Wow.
哇。 - Omniverse works.
Omniverse 有效。 - Olaf, how are you?
Olaf，你好嗎？ - I’m so happy now that I’m meeting you.
我現在見到你好開心。 - I know, because I gave you your computer.
我知道，因為你的電腦是我給你的。 - Jetson.
Jetson。 - What’s that?
那是什麼？ - Well, it’s in your tummy.
嗯，它在你的肚子裡。 - That’s going to be amazing.
那一定很厲害。 - And you learn how to walk inside Omniverse.
而且你是在 Omniverse 裡面學會走路的。 - I love walking.
我愛走路。 - This is so much better than riding on a reindeer gazing up at a beautiful sky.
這比坐在馴鹿上仰望美麗的天空好多了。 - And it was because of physics, using this Newton solver that runs on top of NVIDIA Warp, that we joined together.
正是因為物理學，使用這個在 NVIDIA Warp 上執行的 Newton 求解器，讓我們結合在一起。 - And it was the technology that was completely developed with Disney and with DeepMind that made it possible for you to be able to adapt to the physical world.
而且正是這項與 Disney 和 DeepMind 共同開發的技術，讓你能夠適應物理世界。 - Check that out.
看看那個。 - I used to not to say that.
我以前不會這樣說。 - That’s how smart you are.
你就是這麼聰明。 - I’m a snowman, not a snowclopedia.
我是雪人，不是雪科全書。 - Could you imagine this?
你們能想像嗎？ - The future of Disneyland?
Disneyland 的未來？ - All these robots, all these characters wandering around.
所有這些機器人、所有這些角色到處走動。 - Oh!
噢！ - You know, I have to admit though.
你知道，不過我必須承認。 - I thought you were going to be taller.
我以為你會更高一點。 - I’ve never seen such a short snowman, to be honest.
老實說，我從沒見過這麼矮的雪人。 - Nope.
才不是。 - Hey, tell you what, you want to help me out?
嘿，跟你說，你想幫我一個忙嗎？ - Hooray!
萬歲！ - Okay.
好的。
總結
重點回顧
- Usually, usually I close the keynote by telling you what I told you.
通常，通常我會在結束主題演講時告訴你們我講了些什麼。 - We talked about inference inflection.
我們談到了推論的轉折點。 - We talked about the AI factory.
我們談到了 AI 工廠。 - We talked about the OpenClaw agent revolution that’s happening.
我們談到了正在發生的 OpenClaw 自主式代理革命。 - And of course, we talked about physical intelligence.
當然，我們也談到了物理智慧。 - And that’s what I’m going to talk about today.
這就是我今天要談的內容。 - We’re going to talk about physical AI and robotics.
我們要來談談物理 AI 和機器人。
GTC 閉幕表演
- But tell you what, why don’t we get some friends to help us close it out?
但你知道嗎，何不找些朋友來幫我們做個結尾？ - Of course!
當然！ - All right, play it.
好的，播放吧。 - Come on.
來吧。 - Terminating simulation.
終止模擬。 - Hello?
哈囉？ - Anybody here?
有人在嗎？ - I’m here.
我在這裡。 - Coming alive, agents learning how to drive.
活了過來，代理正在學習如何駕駛。 - From open models to robots too, now we’ll break it all down for you.
從開放模型到機器人，現在我們為你一一拆解。 - Compute exploded, what we saw, from CNNs to OpenClaw.
算力爆炸性成長，我們所見的一切，從 CNN 到 OpenClaw。 - Agents working across the land, but they need the power to meet demand.
代理在各地運作，但它們需要算力來滿足需求。 - So we solved the problem, it was brilliant, we multiplied compute by 40 million.
所以我們解決了這個問題，非常精彩，我們把算力乘以了 4000 萬倍。 - Once upon an AI time, training was the paradigm.
很久很久以前的 AI 時代，訓練是主流範式。 - Sure it taught the models how, but inference runs the whole world now.
訓練確實教會了模型，但現在推論才是驅動整個世界的力量。 - There shows us who’s the boss, at 35 times less the cost.
這讓我們知道誰才是老大，成本降低了 35 倍。 - Blackwell makes the token sing, NVIDIA the inference king.
Blackwell 讓 token 歌唱，NVIDIA 是推論之王。 - Our factories once took years, vendors pulling racks and gears.
我們的工廠曾經需要數年時間，供應商搬運機架和設備。 - Built up slowly, piece by piece, no clear way to scale this beast.
緩慢地一塊一塊建起來，沒有明確的方法來擴展這頭巨獸。 - DSX and Dynamo know what to do, turning power into revenue.
DSX 和 Dynamo 知道該怎麼做，把電力轉化為營收。 - Agents used to wait and see, now act autonomously.
代理過去只會等待觀望，現在能夠自主行動。 - From sim to streets, now watch them drive.
從模擬到街道，現在看它們上路。 - Throw your hands up for physical AI.
為物理 AI 舉起雙手歡呼吧。 - Industrial age, built what came before.
工業時代，打造了過去的一切。 - Now we’re built for AI even more.
現在我們為 AI 打造更多。 - Vera Rubin plus Groq, make the inference splash.
Vera Rubin 加上 Groq，在推論領域掀起波瀾。 - Put them together, now it’s raining cash.
把它們結合在一起，現在是財源滾滾。 - We build new architecture every year.
我們每年都建構新的架構。 - Cause claws keep yelling more tokens here.
因為大家不斷喊著要更多 token。 - The AI stacks for all to make.
AI 堆疊讓所有人都能打造。 - So let us all eat five layer cake.
所以讓我們一起享用五層蛋糕吧。 - The moment’s bright, the path is clear.
此刻光明燦爛，道路清晰明確。 - Cause open models led us here.
因為開放模型帶領我們走到了這裡。 - When data’s missing, there’s no dispute.
當資料不足時，毫無爭議。 - We just generate more with compute.
我們就用算力生成更多資料。 - Robots learning without flaw.
機器人完美地學習著。 - Fueling the force, scaling laws.
驅動這股力量的，是縮放定律。 - The future’s here, won’t you come and see?
未來已經到來，你要不要來看看？ - Welcome all to GTC.
歡迎大家來到 GTC。
閉幕致詞
- We’re here to help.
我們隨時準備提供協助。 - Alright, have a great GTC!
好的，祝大家有個美好的 GTC！ - Wave.
揮手。 - Thank you everybody.
感謝大家。