Name: 英偉達 2024 年人工智能大會：16 分鐘內揭曉一切 (Nvidia 2024 AI Event: Everything Revealed in 16 Minutes)
Uploaded: 2024-06-03T07:06:30.000Z
Duration: 16 min

I hope you realize this is not a concert.

我希望你們明白，這不是一場音樂會。

You have arrived at a developer's conference.

There will be a lot of science described, algorithms, computer architecture, mathematics.

將介紹大量科學知識、算法、計算機結構和數學。

人們認為我們製造 GPU，而我們確實是這樣做的。

But GPUs don't look the way they used to.

And so you could see, I can see, there's a small line between two dies.

你可以看到，我也可以看到，兩個模具之間有一條小線。

This is the first time two dies have been abutted like this together in such a way that the two dies think it's one chip.

這是第一次有兩個模具像這樣對接在一起，以至於兩個模具都認為這是一個芯片。

There's 10 terabytes of data between it, 10 terabytes per second, so that these two sides of the Blackwell chip have no clue which side they're on.

在它之間有 10 TB 的數據，每秒 10 TB，所以布萊克韋爾芯片的兩邊根本不知道自己在哪一邊。

There's no memory locality issues, no cache issues.

沒有內存定位問題，也沒有緩存問題。

It's just one giant chip, and it goes into two types of systems.

它只是一個巨大的芯片，可用於兩類系統。

The first one is form-fit function compatible to Hopper.

第一種是與 Hopper 兼容的外形功能。

And so you slide on Hopper, and you push in Blackwell.

於是你滑動霍普，推入布萊克韋爾。

That's the reason why one of the challenges of ramping is going to be so efficient.

There are installations of Hoppers all over the world, and they could be, you know, the same infrastructure, same design.

世界各地都安裝有 Hoppers，它們可能是，你知道，相同的基礎設施，相同的設計。

The power, the electricity, the thermals, the software, identical, push it right back.

動力、電力、熱能、軟件，完全相同，把它推回去。

And so this is a Hopper version for the current HGX configuration.

是以，這是當前 HGX 配置的 Hopper 版本。

And this is what the second Hopper looks like this.

This is a fully functioning board, and I'll just be careful here.

這是一個功能齊全的董事會，我在這裡會小心謹慎。

This right here is, I don't know, $10 billion?

It gets cheaper after that, so any customers in the audience, it's okay.

之後會越來越便宜，所以觀眾中的任何顧客都可以放心。

The Gray CPU has a super fast chip-to-chip link.

Gray CPU 擁有超快的芯片間鏈路。

What's amazing is this computer is the first of its kind where this much computation, first of all, fits into this small of a place.

令人驚歎的是，這臺計算機是同類產品中第一臺能在這麼小的地方進行如此大量計算的計算機。

They feel like they're just one big happy family working on one application together.

他們覺得自己就像一個快樂的大家庭，一起為一個應用程序工作。

We created a processor for the generative AI era, and one of the most important parts of it is content token generation.

我們為人工智能生成時代創建了一個處理器，其中最重要的部分就是內容令牌生成。

The rate at which we're advancing computing is insane, and it's still not fast enough, so we built another chip.

我們的計算技術發展速度太快了，但還是不夠快，所以我們又造了一個芯片。

這款芯片是一款令人難以置信的芯片。

It's almost the size of Hopper all by itself.

This switch chip has four NVLinks in it, each 1.8 terabytes per second, and it has computation in it, as I mentioned.

這個交換芯片中有四個 NVLink，每個每秒 1.8 TB，而且正如我提到的，它還帶有計算功能。

If we were to build such a chip, we can have every single GPU talk to every other GPU at full speed at the same time.

如果我們能製造出這樣的芯片，我們就能讓每一個 GPU 同時與其他 GPU 進行全速對話。

You can build a system that looks like this.

Just so you know, there are only a couple, two, three Exaflops machines on the planet as we speak, and so this is an Exaflops AI system in one single rack.

你要知道，目前地球上只有幾臺、兩臺、三臺 Exaflops 機器，而這就是一個機架上的 Exaflops AI 系統。

I want to thank some partners that are joining us in this.

我要感謝與我們一起參與這項工作的一些合作伙伴。

They're going to build the first GPU with secure AI.

他們將打造第一個具有安全人工智能的 GPU。

They're building out a 222 Exaflops system for CUDA-accelerating SageMaker AI, for CUDA-accelerating Bedrock AI.

他們正在為 CUDA 加速的 SageMaker AI 和 CUDA 加速的 Bedrock AI 構建一個 222 Exaflops 系統。

Amazon Robotics is working with us using NVIDIA Omniverse and Isaac Sim.

亞馬遜機器人公司正在使用英偉達™（NVIDIA®）Omniverse 和 Isaac Sim 與我們合作。

AWS Health has NVIDIA Health integrated into it, so AWS has really leaned into accelerated computing.

AWS Health 中集成了英偉達™（NVIDIA®）Health，是以 AWS 真的向加速計算領域傾斜了。

谷歌正在為布萊克韋爾做準備。

GCP already has A100s, H100s, T4s, L4s, a whole fleet of NVIDIA CUDA GPUs, and they recently announced the GEMMA model that runs across all of it.

GCP 已經擁有 A100、H100、T4、L4 和一整套英偉達™（NVIDIA®）CUDA GPU，他們最近還發布了跨所有 GPU 運行的 GEMMA 模型。

We're working to optimize and accelerate every aspect of GCP.

我們正在努力優化和加速 GCP 的各個方面。

We're accelerating Dataproc, which is for data processing, their data processing engine, JAX, XLA, Vertex AI, and MuJoCo for robotics, so we're working with Google and GCP across a whole bunch of initiatives.

我們正在加速 Dataproc（用於數據處理）、他們的數據處理引擎、JAX、XLA、Vertex AI 和用於機器人技術的 MuJoCo，是以我們正在與谷歌和 GCP 開展一系列合作。

甲骨文正在為布萊克韋爾做準備。

Oracle is a great partner of ours for NVIDIA DGX Cloud, and we're also working together to accelerate something that's really important to a lot of companies, Oracle Database.

甲骨文是英偉達 DGX Cloud 的最佳合作伙伴，我們還在合作加速對許多公司非常重要的 Oracle 數據庫。

Microsoft is accelerating, and Microsoft is gearing up for Blackwell.

微軟正在加速發展，微軟正在為布萊克韋爾做好準備。

Microsoft and NVIDIA has a wide-ranging partnership.

微軟與英偉達™（NVIDIA®）建立了廣泛的合作關係。

We're accelerating CUDA-accelerating all kinds of services when you chat, obviously, and AI services that are in Microsoft Azure.

當你哈拉時，我們正在加速 CUDA--加速各種服務，很明顯，還有 Microsoft Azure 中的人工智能服務。

It's very, very likely NVIDIA is in the back doing the inference and the token generation.

英偉達很有可能在後面進行推理和令牌生成。

They built the largest NVIDIA InfiniBand supercomputer, basically a digital twin of ours or a physical twin of ours.

他們建造了最大的英偉達InfiniBand超級計算機，基本上是我們的數字孿生或物理孿生。

We're bringing the NVIDIA ecosystem to Azure, NVIDIA DGX Cloud to Azure, NVIDIA Omniverse is now hosted in Azure, NVIDIA Healthcare is in Azure, and all of it is deeply integrated and deeply connected with Microsoft Fabric.

我們正在將英偉達生態系統引入 Azure，將英偉達 DGX Cloud 引入 Azure，英偉達 Omniverse 現在託管在 Azure，英偉達醫療保健也在 Azure，而且所有這些都與 Microsoft Fabric 深度集成、深度連接。

And now, it's a pre-trained model, so it's pretty clever, and it is packaged and optimized to run across NVIDIA's installed base, which is very, very large.

現在，它是一個預先訓練好的模型，所以非常聰明，而且經過打包和優化，可以在英偉達龐大的安裝基礎上運行。

You have all these pre-trained, state-of-the-art open-source models.

你擁有所有這些預先訓練好的、最先進的開源模型。

它們可能來自我們的合作伙伴。

It could be created by us, like NVIDIA Moment.

它可以由我們創建，比如 NVIDIA Moment。

It is packaged up with all of its dependencies.

So CUDA, the right version, CUDNN, the right version, TensorRT, LLM, distributing across the multiple GPUs, Trident inference server, all completely packaged together.

是以，CUDA（正確的版本）、CUDNN（正確的版本）、TensorRT、LLM（在多個 GPU 上分佈）、Trident 推理服務器，都完全打包在一起。

It's optimized, depending on whether you have a single GPU, multi-GPU, or multi-node of GPUs, it's optimized for that, and it's connected up with APIs that are simple to use.

根據你是使用單 GPU、多 GPU 還是多節點 GPU，它都進行了優化，並與簡單易用的 API 相連接。

These packages, incredible bodies of software, will be optimized and packaged, and we'll put it on a website, and you can download it, you can take it with you, you can run it in any cloud, you can run it in your own data center, you can run it in workstations if it fit, and all you have to do is come to ai.nvidia.com.

這些套裝軟體，令人難以置信的軟件體，將被優化和打包，我們將把它放在一個網站上，你可以下載它，你可以隨身攜帶它，你可以在任何雲中運行它，你可以在你自己的數據中心運行它，如果合適的話，你可以在工作站中運行它，你所要做的就是訪問 ai.nvidia.com。

We call it NVIDIA Inference Microservice, but inside the company, we all call it NIMS.

我們稱之為 NVIDIA Inference Microservice，但在公司內部，我們都稱之為 NIMS。

We have a service called NIMO Microservice that helps you curate the data, preparing the data so that you could teach this, onboard this AI, you fine-tune them, and then you guardrail it, you can even evaluate the answer, evaluate its performance against other examples.

我們有一項名為 "NIMO 微服務 "的服務，可以幫助你整理數據、準備數據，這樣你就可以教授人工智能，讓人工智能上崗，對其進行微調，然後對其進行護欄，甚至可以評估答案，根據其他例子評估其性能。

是以，我們實際上是一家人工智能代工廠。

We will do for you and the industry on AI what TSMC does for us building chips.

在人工智能領域，我們將像臺積電為我們製造芯片那樣，為您和業界做出貢獻。

And so we go to it with our, go to TSMC with our big ideas, they manufacture it, and we take it with us.

是以，我們帶著我們的遠大理想去找臺積電，他們生產，我們帶走。

And so exactly the same thing here, AI foundry, and the three pillars are the NIMS, NIMO Microservice, and DGX Cloud.

人工智能代工廠也是如此，三大支柱是 NIMS、NIMO 微服務和 DGX 雲。

We're announcing that NVIDIA AI foundry is working with some of the world's great companies.

我們宣佈英偉達™（NVIDIA®）人工智能代工廠正在與世界上一些偉大的公司合作。

SAP generates 87% of the world's global commerce.

從根本上說，世界是靠 SAP 運行的。

NVIDIA and SAP are building SAP Jewel co-pilots using NVIDIA NIMO and DGX Cloud.

英偉達™（NVIDIA®）和SAP正在利用英偉達™（NVIDIA®）NIMO和DGX Cloud打造SAP Jewel協同駕駛儀。

ServiceNow, they run, 85% of the world's Fortune 500 companies run their people and customer service operations on ServiceNow.

全球 85% 的財富 500 強企業都在 ServiceNow 上運行其員工和客戶服務業務。

And they're using NVIDIA AI foundry to build ServiceNow assist virtual assistants.

他們正在使用英偉達™（NVIDIA®）人工智能代工廠構建 ServiceNow 輔助虛擬助手。

Hundreds of exabytes of data, over 10,000 companies.

數百億字節的數據，超過 10,000 家公司。

NVIDIA AI foundry is working with them, helping them build their Gaia generative AI agent.

英偉達™（NVIDIA®）人工智能代工廠正在與他們合作，幫助他們建立 Gaia 生成式人工智能代理。

Snowflake is a company that stores the world's digital warehouse in the cloud and serves over 3 billion queries a day for 10,000 enterprise customers.

Snowflake 是一家在雲中存儲全球數字倉庫的公司，每天為 10,000 家企業客戶提供超過 30 億次查詢服務。

Snowflake is working with NVIDIA AI foundry to build co-pilots with NVIDIA NIMO and NIMS.

雪花公司正在與英偉達™（NVIDIA®）人工智能代工廠合作，利用英偉達™（NVIDIA®）NIMO 和 NIMS 構建協同飛行器。

NetApp, nearly half of the files in the world are stored on-prem on NetApp.

全球近一半的文件都存儲在 NetApp 公司內部。

NVIDIA AI foundry is helping them build chatbots and co-pilots like those vector databases and retrievers with NVIDIA NIMO and NIMS.

英偉達™（NVIDIA®）人工智能代工廠正在幫助他們利用英偉達™（NVIDIA®）NIMO 和 NIMS 構建哈拉機器人和輔助機器人，如矢量數據庫和檢索器。

And we have a great partnership with Dell.

我們與戴爾公司建立了良好的合作關係。

Everybody who is building these chatbots and generative AI, when you're ready to run it, you're going to need an AI factory.

每個正在構建哈拉機器人和生成式人工智能的人，當你準備好運行它時，都需要一個人工智能工廠。

And nobody is better at building end-to-end systems of very large scale for the enterprise than Dell.

在為企業構建端到端大規模系統方面，沒有人比戴爾更勝一籌。