NVIDIA Keynote at COMPUTEX 2023

"I am a translator, I am a healer, I am a visionary, I am a navigator, I am a creator, I am a helper, I am AI",(Vedio,16:00 - 19:14)

接續便接續進入了主要環節，講述到了六年前相比目前的光追(Ray Tracing)進展，從需要一段時間，變為今日的Real Time。第一個出場的產品便是RTX 4060TI

NVIDIA ACE For Game，並配合影片講述，遊戲如何Real Time呈現，並且生成角色特性，以及其BackStory，使其可以更加自然的融入在遊戲中

再來提到了電腦的進展，從1964年的IBM System 360,

"This computer not only revolution computing, it revolutionize that the thinking of the computing industry."

這個架構持續了60幾年至今

3 Trends:

1. CPU Scaling is ending. The ability of get 10 times more performance every 5 year has ended, by the same cost and without increasing of power.

Deep Learning, Data center and generated AI

2. Data Center scale problem, today, data center is a computer
3. The celerity computing is multi-domain.

Computing virtuous cycle ,the loop by the last 40 years computer companies.

For example, a 10 million dollar server,

"That is why the reason that the people say that the GPU servers are so expensive. However, the GPU server is no longer than computer, the computer is a data center. Your goal is to build a casetify data center, not casetify server. Back to the old day that computer was the server, that will be the reasonable thing to do, but today, the computer is a data center."

Before	After

"We want dense computer, not big ones. Almost everybody is power limited, if your goal is to get the work done, you don't care how, and this is the work you want to get done, iso work."

Now: The more you buy, the more you save.

Data Center TCO(總持有成本)

Lifecycle optimization, it benefit everybody.

Generative AI

We are now reach the tipping point of Generative AI

H100

H100 is a full production, manufaction by company all over Taiwan, used in cloud everywhere.

Let's see a video how H100 produced(47:21 - 48:18)

A first computer in the world that a transformer AI engine in it.

A whole new way to use AI for developing software. We reinvented the used for GPU. Every data center had been reinvented.

DGX

A giant GPU Computer. If a data center is a computer, than the networking defines the data center, that is incredibly good a position, and seems than we done so many things together that I want to show you some really really amazing word today.

Every two years, we took a giant leap forward, but we realize we need more even that. And which is the reason why we connected GPU to other GPU called NVLink, build one giant GPU, and we connected those GPU together, using infiniband larger scale computer. The ability for us to drive the processor, and extend the scale of computing, made it possible for the AI　research organization, that communicated to advance AI to an incredible way.

We just keep pushing and pushing and pushing. Hopper when it production, August of last year, August 2022. 2024, which is the next year, we got hopper next. Last year we got quantum, and next year we have quantum next.

Every two years we've get giantly forward, and we expend it to be giantly as well.

This is a new computer industry. Software is no longer programmed just by computer engineers, software is programmed by computer engineers working with AI super computers.These new AI super computers, are new type of factory, it is very logical that the car industry has factories they build that you can see cars; it is very logical that computer industry has computer factories you build things you can see. Computers, in the future, every single major company, will also have AI factories. And you will build and produced, your company intelligence.

We are intelligence producers already, and everything will have factory in this way.

This translate to your throughput, this translate to your scale, and all of you will build it in the way to a vary vary goot ti ceo.

Generative AI

In five years, we improve computer graphics by one thousand times. That is 1 millions in 10th. What would you do if the computer is 1 million times faster?

Deep learning and chatGPT, the able to learn the giant amount of data, and recongnize the pattern and the relationship accross a large sequence. And using transformor to predict the next word, large language model were created.

We can learn the structure of text, sound, images, the structure all of this, physics, proting, DNA, chemical, anything that has structure, we can learn that language, leatrn its language.

So many types of information, can now be transformed. For the very first time in history we has a software technology, that is able to understand the representation of many information of many modeless.

We can apply so many industry to so many different feels that were impossible before. This is the reason why that everybody are so excited.

AI展示(1:00:06-1:04:30)

文字輸入	影片生成

音樂生成

文字輸入	KTV生成

Video showing(1:05:23-1:06:28): working with generative AI

1. It is able to understand information of more than just text and numbers. It can now understand multimodal, which was the reason why this computing revolution can impact every industry.

2. Because these computers, doesn't care how you programmed at it. It will tried to understand what's your mean. Because it has the incredible large language model capability. And so the programming barier is incredibly low. We have closed the digital devide, everyone is a programmer now. Just have to say something to the computer.

3. This computer, not only to do the amazing things for the future, it can do the amazing things for every single application of the previous era. Which is the reason why all of these API had been connected to the Windows application and to the browser, power point, the words, every application that exists WILL be better because of AI. You don't have just AI, this generation, this computing era, does not need new applications, it can succeed with old applications. And it can have new applications.

The rate of progress, because it's so easy to use, it's the reason why it growing so fast. This is going to touch literally every single industry, and that the core, just adding in every single computing era, it needs a new computing approach. And this particular, in this particular era that computing approach is an softwarely computing, and it has been completely being reinvent in the ground up.

And that is why grace hopper been

Grace Hopper, GH200

This is the world first solarity processor, that also has a giant memory. It has almost 600 GB of memory that coherence between CPU and GPU. So that CPU can reference the memory, that GPU can reference the memory, unnecessary, any unnecessary the copied and back formed, could be avoided. The amazing amount of high speed of memory, lets the GPU working on a very very large datasets. This is a computer, this is not a chip. Practically entire computer is on here. Use low power DDR memory, just like your cellphone. It took us so many years to build.

Performance comparing:

The performance speedup, I can't wait to show it to you, this is revolutionize the entire industry what is a higher computing industries in the world of cource.

So literally in the course of 10 years, the computing problem of deep learning, increase by five thousand times for the software, and three million times for the dataset.

This is going to made a big big contribution, however 600 GB is still not enough. We need a lot more.

First, we put GH200 into computers, the second thing we going to do is to connect 8 of these together, using NVLink, by using 3 of these NVLink switches as a pipe, and we connected 32 of these together with another of switches, and inorder to build this. Have 144 TB memory connected,not seperated, that every GPU could see.

Showing how DGX GH200 construction(1:18:12-1:18:30)

One giant GPU, one DGX GH200.

Companies

What if the GH200 with generative AI in our life(Video 1:22:46-1:24:20)

It could also run the 5G stack. Then the telecommunication netwrk can also become a computing platform like the cloud data centers. Every single data center in the future could be intelligence.

NVIDIA MGX

It's a open modulor server design decertification , and the design for salarity computing. Most server design are for the general purpose computers.

Various type severs in MGX

The data center is the computer. The Network defines the data center.

Announcing STPECTRUM-X

1. Adopted Routing
2. Congestion Control

Worlds most valuable companies need custom creative AI

To help the industries build custom language models, not every body can use the language models in available in a public service, as some customers need language model that is highly especiallize for their particular modelic.

This help you to train your own AI model:

CPU v.s. NVIDIA　AI GPU

Available on the cloud companies

NVIDIA Omniverse

AI have a digital twins, for heavy industries.

How AI transfer?

Nothing was art, everything is simulation. Is it amazing? (Vedio: 1:50:13-1:52:39)

NVIDIA omniverse (Vedio: 1:53:46-1:55:26)

Everything is ray trace, no art is necessary, you bring everything entire captipn to omniverse, open your browser, bring your factory in,no art is necessary, the light in just the light in does, physics just in the physics does. And multiple users, as many as you like, can enter the omniverse on the same time, and work together.

The world biggest advertise company, WPP, is parting with NVIDIA to build a content generation enging, based on omniverse and generative AI.

How WPP use these technologies(vedio:　1:58:59－2:00:19)

How we could use these technologies in Taiwan,Semiconductor industry(vedio:　2:01:22-2:03:40)

Whole factories are omniverse, are whole digital. Imaging you have digital information in your hands, what can you do with it? Almost everything. And so this is the thing that really exciting.

NVIDIA Robot platform, ISAAC AMR

This is the blue print for ISAAC AMR (vedio:2:06:02-2:07:37)

IT　insdustry ecosystem

Thank all of you for coming today. I talked about many things, it's be a long time since I've seen you. So I had so muxh to tell you, it was too much.

I told you several things, I told you that we are going to two simultaneous computing industry transition, is already computing generative AI; two, this form of computing, is not like the traditional general computing, it is full stack, it is data center scale, because the data center is a computer, and it is domain specific. For every domain that you want to going to, every industry you going to, you need the software full stack. And if you have the software full stack, then you expandly utilization of your machine.

Lastly, we would like to extend AI to the world heavy industry, the largest instry in the world.