No One is Talking About What Just Happened

Just when you think the world couldn't get any weirder than it is now, this happens and no one seems to be

talking

about it, let's talk about the next wave of robotics, the next wave of AI, robotics, physical AI, so far, everything the AI. What we've talked about is that data from a computer goes into a computer, we take all the data and put it into a system like dgx, we compress it into a big language model, billions of tokens become billions of parameters, these billions of parameters become your AI. So I described in very simple terms essentially

what

just

happened

in large language models, except that the GPT chat moment for robotics may be right around the corner, so we've been building end-to-end systems for robotics for some time.

I'm super very proud of the work we have the dgx AI system we have the bottom system called agx for autonomous systems the world's first robotic processor when we first built this, people are

what

are you guys building? It's an s, so it's a ride it's designed for. It will be very low power but is designed for high speed sensor processing and artificial intelligence. Things are getting crazy out there Fanny maybe they were right they were right they were right I can't believe it it's actually happening it's actually happening I can't believe it It's actually happening I guess I'm not used to Killer Robots chasing me around a mall in the middle of the night.

Okay, let's delve a little deeper into this world of artificial intelligence and the machinery behind it. Let's think of AI as This ultra-intelligent robot brain is learning at crazy speeds and tackling things we once thought only the human brain could tackle. You name it, AI is doing it, but here's the problem: the smarter these robot brains get, the more blurred the line between controlling them and being controlled by now becomes hardware, real metal and microchips in These AI brains live on what we call GPUs, which have traditionally been used for video games; however, currently, gpus or graphics processing units are what is used to train Ai and the leading company that manufactures these gpus.

Is it NVIDIA, have you heard of Mo's law which says that technology gets twice as good every 2 years or so, but as these machines get stronger historically we've always feared that what we can't control will take over. operation and H's reflection on the creation of the atomic bomb?, he said quote I have become death, the destroyer of worlds and that fits perfectly with what is happening in the world of artificial intelligence. What you are about to see is a presentation from the CEO of Nvidia, who makes all the GPUs that power all the AI ​​on Earth and what he reveals in this presentation shows how far we have come in the last few years.

The pace at which we advance. Computer science is crazy. Over the course of the last 8 years, we have increased computing 1,000 times in 8 years. 1000 times, remember, in the old days of Mo's law, it was 10 times every 5 years 100 times every 10 years in the middle of the PC Revolution 100 times every 10 years in the last eight years we have gone 1000 times, we have two more Years to go, the pace at which we advance Computing is crazy and still not fast enough, so we build another chip. This is the world's most advanced GPU in production today. This is Hopper. This is hopper.

Hopper changed the world. This is Blackwell. 208 billion transistors, so you can see it. I can see that there is a small line between two dies. This is the first time that two dyes have been joined together in this way in such a way that the two chips think it is one chip. There are 10 terabytes of data between them 10 terabytes per second, so these two these two sides of the Blackwell chip have no idea which side they are on, there are no memory locality problems or cash problems, it's

just

a giant chip when we they said that Blackwell's ambitions were beyond the limits of physics uh the engineer said so what and this is what

happened

and this is the Blackwell chip and it goes into two types of systems, the first has a tuning function Hopper compatible way, so you slide on Hopper and push on Blackwell, that is the reason why one of the ramp challenges will be so efficient, there are Hoppers installations all over the world and the same infrastructure, the same design, power, electricity, thermals, identical software, push it back, and so on. it's a hopper version for the current hgx setup, the second Hopper looks like this now, this is a prototype board and this is a fully functional board and just be careful, this right here is, I don't know, 10 billion, the second one is five, it gets Cheaper after that, so the way it will go to production is like this: two black chips and four Blackwell dyes connected to a Grace CPU.

The Grace CPU has super-fast chip-to-chip linking. The amazing thing is that this computer is the first of its kind. where so much computing fits in such a small place, but we need a lot of new functions to push the limits Beyond, if you will, the limits of physics, so one of the things we did was invent another Transformer engine and So, this new Transformer engine, we have a 5th generation MV link, it is now twice as fast as Hopper, but the most important thing is that it has network computing and the reason is that when you have so many different GPUs working together, we have to share our information. between us we have to synchronize and update each other, having extraordinarily fast links and being able to do math directly on the network allows us to essentially amplify ourselves even more, so even though it is 1. terabytes per second, it is effectively higher than that and therefore Therefore, there are many. times higher than Hopper overall compared to Hopper, it is two and a half times the performance of fp8 for per-chip training.

It also has this new format called fp6, so even though the calculation speed is the same, the number of parameters it can store in memory. Now Amplified FP4 effectively doubles performance. This is vitally important for inference. The amount of energy we save. The amount of network bandwidth we save. The amount of time lost we save will be tremendous. The future is generative, which is why we call it. Generative AI, which is why this is a completely new industry, the way we compute is fundamentally different. We built a processor for the era of generative AI and one of the most important parts is the generation of content tokens.

Call. This format is fp4. a lot of computation, 5x token generation, 5x Hopper inference power seems enough, but why stop there? So we would like to have a bigger GPU, even bigger than this one, and we decided to scale it up, so we built another chip, this chip. it's just an amazing chip, we call it mvy link switch, it has 50 billion transistors, it's almost the size of Hopper alone, this switch has four MV links each of 1.8 terabytes per second, what is it for? this chip if we built such? one chip we can make each GPU talk to every other GPU at full speed at the same time, that's crazy and as a result you can build a system that looks like this, this is what a dgx looks like now, remember that I gave it away only 6 years ago. the first djx1 to open IA, that djx by the way was 170 teraflops, that is 0.17 pedoflops, so this is 720, so now it is 720 pedop flops, almost an exaflop for training and the first one exop flops machine in the world in a single rack, just so we know there are only a couple of two exop flops machines on the planet as we speak, so this is an exop flops artificial intelligence system in a single rack, well , let's take a look at the back, so this is what makes it possible.

That's the back of the dgx MV link. 130 terabytes per second pass through the back of that chassis, which is more than the aggregate bandwidth of the Internet. Basically, we could send everything to everyone in a second. Let's talk about the next wave of robotics. next wave of robotic AI Physical AI so far all the AI ​​we have talked about is a computer, the data goes into a computer, we take all the data and put it into a system like dgx, we compress it into a big language model, billions of tokens become billions of parameters, these billions of parameters become your AI, so I just described in very simple terms, essentially, what just happened in large language models, except that the GPT chat time for robotics may be just around the corner so we have been building the end to end systems for robots robotics for some time now I am super proud of the work we have the dgx artificial intelligence system we have the lower system called agx for autonomous systems the world's first robotics processor when we first built this thing people What are you guys building?

It's a chip, it's designed to be very low power, but it's designed for high speed sensor processing and Ai, so if you want to run Transformers on a car or anything that moves, we have the perfect computer for you. called Jetson and that's why the dgx at the top to train the AI, the Jetson is the autonomous processor and in the middle we need another computer, we need a simulation engine that represents the world digitally for the robot so that the robot has a gym to learn how to be a robot, we call that virtual world Omniverse and the computer that runs Omniverse is called ovx and ovx, the computer itself is hosted in the Azure CL cloud, so we basically built these three things in which these three Furthermore, we have algorithms for each one.

Now I'll show you a super example of how Ai and Omniverse are going to work together. The example I'm going to show you is a little crazy but it will be very very close tomorrow is a robotics building this robotics building is called a warehouse inside the robotics building there will be some autonomous systems some of the autonomous systems systems will be called human and some of the autonomous systems are going to be called forklifts and these autonomous systems are going to interact with each other, of course, autonomously and this warehouse will bypass them to keep everyone out of harm's way.

The warehouse is essentially an air traffic controller and every time it sees something happening it will reroute traffic and give new waypoints, just new waypoints to the robots and to the people, and they will know exactly what to do with this warehouse, with this building, you can talk too, of course, you can talk to it, hey, and this is all running in real time, what's up with everything? the robots, all those robots that you were looking at a moment ago, they all run their own autonomous robotic stack, let's talk about robotics, everything that moves will be robotic, there is no doubt that it is safer, it is more convenient and one of the industries bigger it will. be Automotive early next year we will ship at Mercedes and then shortly after we will announce that byd, the world's largest electric vehicle company, is adopting our next generation.

His name is Thor. Thor is designed for Transformer motors. Thor, our next generation. Byd will use the AV computer. The next generation robotics will likely be humanoid robotics. We now have the technology to imagine widespread human robotics in a way that human robotics is probably easier and the reason is that we have a lot more training data. that we can provide the robots because we are built in a very similar way, it could be in video form, it could be in virtual reality form, then we created a gym for it called Isaac reinforcement learning gym, which allows the humanoid robot to learn to adapt. to the physical world and then to an amazing computer the same computer that will go into a robotic car this computer will run inside a human or robot called Thor it is designed for motors Transformer the soul of Nvidia the intersection of the computer graphics physics artificial intelligence all They came to carry at this time the name of that general robotics project 003 I know very well, very well, I think we have some special guests, right?

Hey guys, so I understand that you are powered by Jetson, you are powered by Jetson, little Jetson. robotic computers inside learn to walk in Isaac Sim ladies and gentlemen this this is orange and this is the famous green these are Disney's bdx robots amazing Disney research come on guys let's finish let's go five things where are you going what are you? saying no, it's not time to eat, it's not time to eat. I'll give you a snack in a moment. Let me finish real quick first. A new industrial revolution. Every data center should be accelerated. One trillion dollars in installed data.

The centers will be modernized in the coming years according to the computer of thisrevolution the computer of this generation Generative AI billions of parameters this is what we announce to you today this is Blackwell incredible processors incredible switches mvlink network systems and the system design is a miracle this is Blackwell and this is to me what a GPU looks like in my mind, everything that moves in the future will be robotic, you're not going to be the only one and these robotic systems, whether they're humanoids, self-driving cars, forklifts that manipulate arms. everyone will need one thing Giant stadiums warehouses factories there may be factories that are robotic manufacturing lines that are robotic building cars that are robotic these systems everyone needs one thing they need a platform a digital platform a digital twin platform and we call that Omniverse the operating system From the world of robotics, well, first there were robots, then there were self-replicating robots, there is a growing fear that robots will continue to make the human workforce obsolete before the success of the ubiquitous humanized police robots.

There was a bigger bad boy on the Block. Moose, we are able to give robots superficial responses that mimic human expression. I saw something incredible. I saw three robots completely reprogrammed in 10 minutes we created the robot race the robot race so there is the proof that things are changing faster than most people can imagine, then there is the question what comes after all this, Aristotle once said knowing yourself is the beginning of wisdom, but if AI can think, decide and perhaps one day understand itself better than we do, where does that leave us? ? Let's forget about all the AI-powered gadgets and gadgets for a second and focus on what's about to happen.

As we speak, they are implanting AI into real robots and they are building these things right now with plans to release them to humanity in a few years. This is no longer science fiction, just think about what it was a few years ago in 2021, that was 3 years ago in In that same period of time going forward we will see robots, we are building a world where AI could decide everything from who could getting a loan to who is fit to raise a child, including who is guilty of a crime. It's like opening Pandora's box, which once opened can't be closed again, but what do you think about this?

I want to hear your opinion. Please leave your opinion in the comments section if you haven't already and don't forget to hit the like button and until next time, thank you. for watching and God bless you all