Tesla Reveals The New DOJO Supercomputer!Aug 29, 2023
Tesla's Artificial Intelligence Division has created a very special and very powerful new type of
supercomputerand they call it a
dojo. This is a project that Tesla has been talking about for the past few years, but as of now Dojo is online and growing. In power at an incredible rate it is expected to reach the top 5 most powerful computers in the world early next year the Dojo project is a much bigger business than most people realize and it is about to change everything for Tesla in June 2023 Elon Musk announced that Dojo has been online and running useful tasks in Tesla's data centers for a few months.
This was the first confirmation we had that Dojo is operational and ready for action. This puts an end to speculation that has been circulating about how mature this new system really is. is and elon's words were backed up by information from Tesla's new official AI Twitter account showing that the Dojo hardware will be ready for production in July 2023. This means that Tesla has already started fully stockpiling Dojo chips. functional and to build them into larger systems of racks and cabinets. As supply continues to grow exponentially, Tesla AI predicts that Dojo will expand from its current testing phase to become one of the world's five most powerful
supercomputers by February 2024, reaching more than 30 exafly of computing power and, at From there, Dojo Hardware's production rate will continue to increase in volume, bringing Tesla's total computing power to 100 exaflops by October 2024.
Now you might be wondering what the hell an exaflop is. It is a way of measuring the amount of work a computer can handle in one second. An exaflop represents a quintillion computer operations per second, that is, a one with 18 zeros behind it, it is an unfathomable number on the human scale, so let's put it into a metric that we all understand: money. 100 exaflops of compute is equivalent to 300,000 Nvidia a100s. GPUs These have been the industry standard for some years as the top-tier processing unit in the world's most powerful data centers. These are not the type of chips used for video games.
These A100s are used to create the most advanced software in the world. AI models like gbt chat and midway image generator would not exist without the a100 chip. And that, in turn, makes these GPUs have an incredibly valuable average cost of around ten thousand dollars per unit, so now we can take 100 exaflops, which is 300,000 a100 multiplied by ten thousand dollars and you get three billion of dollars that's what Tesla is creating with Dojo a three billion dollar supercomputer and that's right where
dojowill be next October according to the initial curve on the Tesla graph it seems like they are projecting at least a 100 increase in capacity over a period of one year, so it all sounds very exciting and cool, but what is this Dojo actually going to do?
Why does a car company need such a ridiculously powerful computer? There's a lot to unpack here, just like in a real-world dojo. This computer system is designed to function as a training ground, except instead of karate, the Tesla Dojo will train artificial intelligence, more specifically, this is the new home of Tesla's fully autonomous neural network, so this is the point where we can begin to separate the Dojo from the standard. definition of supercomputer, you see, Dojo is more correctly described as an artificial intelligence training group, but supercomputer is a much more familiar association to the average person, so we start with that to establish the fundamentals of these training groups which are traditionally made up of giant full cabinets.
With GPUs running, the graphics processor we've all been using for decades turned out to be particularly well-suited for the kind of computation that training neural networks demands, so companies like Nvidia and am D started making them much bulkier and smaller versions. powerful of their existing designs, but Dojo comes to the game with a completely new approach. Dojo is a custom hardware platform that was designed from the ground up by Tesla's AI division exclusively for use in training its latest full self-control based computer vision video. driving networks, the goal is to create a digital duplicate of the human visual cortex and brain function and then use it to drive a car autonomously.
This involves processing large amounts of visual data which, in this case, are videos captured by the vehicle's cameras, all that information needs to be translated billions of frames of digital videos into a language that the AI model can understand. This is called tagging and it's exactly what it sounds like, as if they're simply assigning a designation to a group of pixels so the AI doesn't know what it's looking at. The more labels the network has to extract, the better it can recognize patterns and make associations. In the past, Tesla has had humans do this labeling work, but that is obviously not sustainable to increase FSD capabilities by orders of magnitude over time.
We would have to have every human being on Earth working on Tesla, so to be successful they need to automate and that automation comes in the form of computer power which has now taken the form of the Dojo. Tesla's work with Dojo is strikingly similar to what Apple has been doing with its own computer systems in both philosophy and technology. Apple discovered a long time ago that it was a really effective strategy to create software and hardware specifically designed to work together, resulting in a more efficient, higher-performing device. This is something that Apple has fully realized with its new M1 and M2 powered computers.
They've replaced the Intel processors with their own custom chip designed specifically to run Apple software and they're doing it in a completely different way than any other computer. In the market, what really sets Dojo apart from the rest of the AI training industry is its move away from GPU hardware. Dojo exists at its basic level as something called a system on a chip, which is a complete computer assembled on a single piece of silicon and This is the exact same architecture that Apple used to create the M1, this method allows for a spectacular level of efficiency because Instead of having all these PCI ports, cables, motherboards and such connected together, now all the necessary components live in the same small square of semiconductor material and the more power you need, the bigger you make that piece of silicon and the more cores processor plug it in, you can see that with Apple's M1 M1 Pro and M1 Max chips, the M1 doesn't fit inside an iPad or Mac.
MacBook Air, while the M1 Max is the size of a MacBook Pro, the dojo chip It's about palm-sized, which is much smaller than an a100 GPU, but Dojo isn't actually supposed to exist as a single unit. it becomes functional at the tile level, which is the point where multiple chips are merged to function as a single system and again, this is something Apple is also doing with its new Mac Studio units, its top-tier M1 Ultra chip are just two M1 Max chips fused together. together to create an enormously powerful computer with the dojo tile Tesla has fused 25 Dojo chips to create a unified computing system and each tile contains all the hardware necessary for electrical cooling and data transfer.
It is a self-contained computer itself made up of 25 smaller computers then go up a level and integrate six tiles into a single rack unit and then to make a cabinet they integrate two of the racks into a single case. The amazing thing about this system on a chip architecture is the level of efficiency it can achieve. It is achieved with minimal need for power or cooling, so looking back at Apple, its Mac Studio with the M1 Ultra or now the M2 Ultra, these are the most powerful computers you can buy, but they are housed in these tiny square boxes that can be easily placed on top. on your desk they take up very little space and only need a basic fan system for cooling.
It's amazing really, just the power supply on a traditional desktop PC would be half the size of an Ultra powered Mac Studio. Hopefully that will give you a more tangible experience. idea of how Dojo is operating the same thing that is happening inside your new Macbook is happening inside Dojo on a large scale, so now what does it all mean? What does it mean for this Dojo thing to change for Tesla? Well, number one is obviously that they will be able to quickly add a lot of computing power to their AI training program at a relatively low cost to the company.
Obviously any new product will have very high overhead costs at the start of production, but the more Dojo chips and tiles there are. that are produced the more affordable they become, this also means that Tesla will not go to the same Marketplace and compete to buy the same Nvidia chips as everyone else in its industry, as the power of AI continues to grow, as does demand. of a100 and h100 level processors will be difficult to acquire in large quantities and the price will reflect the demand for Nvidia's new h100 GPU, which currently costs 40,000 per unit and this inflated value placed on AI training power It's something Tesla could leverage in the future to create an entirely new business model with its existing AI division.
Elon Musk has said that this first version of Dojo is designed specifically for Tesla's computer vision video tagging, which is exactly what they need for FSD and later the humanoid Tesla. bot, so Dojo won't be particularly useful for anything beyond that, but Elon says future versions of the dojo system will be more tailored to general-purpose AI training, so it could be tailored to language models or algorithms. social networks or anything else. Basically, once Tesla gets their own Dojo system up to a level where it delivers all the computing power they need, then every additional Dojo system they build will become an asset that can be monetized.
Elon sees this working exactly the same. As something like Amazon Web Service or Microsoft Azure Tesla will simply rent its excess computing power to anyone who needs it and there will be a staggering amount of need for this service in the coming years, this type of business model is as lucrative as it gets. that Amazon's web service is a spectacularly profitable division. This is the reason why Jeff Bezos became the richest man in the world. It's the reason why Amazon can sell all of this so cheap and deliver it so fast that it finances itself by simply renting its web service with additional server capacity.
It started because Amazon only really needed its maximum server capacity for peak periods like Black Friday, the rest of the year it just sat there doing nothing until they came up with the idea of renting it out. Dojo can do exactly the same for Tesla and this is what we call a game changer. Don't forget to like this video today if you liked it. It is very important to get our content to more people. If you enjoy the con, you'll probably also enjoy our weekly content. Newsletter, so sign up with the link below at the
A big thank you to all of our Patreon supporters listed above. Now they help us create the best content possible and we really appreciate it. Thanks for looking and us. See you next time
If you have any copyright issue, please Contact