Tesla has now unveiled its newest supercomputer, already the fifth most powerful computer in the world. And they won’t be stopping there – another supercomputer is coming which will be more powerful. That will be the much-talked-about Dojo Supercomputer. Currently, this new computer will be using neural nets to power Tesla’s Autopilot and self-driving AI.
Tesla and Artificial Intelligence
In recent years, Tesla has done a lot of work on improving the computing power in their vehicles. This includes the power inside the vehicle’s infotainment system as well as the power that the computers inside the vehicle use. Now a self-driving software, be it completely autonomous or partially, needs a lot of computing power. With the number of variables that a computer needs to factor in, the required computing power goes sky-high.
And that’s not all. Running self-driving software is one thing, but training your system is a whole other game. You need neural nets to make sure the system learns about the decisions it needs to take and you also need to accommodate the humongous amount of data coming in. And this data isn’t just coming from the sensory systems from your vehicle. Tesla connects all its vehicles to a mothership, and the data coming from the entire fleet of its vehicles gets accumulated.
The Dojo Project
Tesla CEO Elon Musk has been continually teasing the public about this Dojo project. This supercomputer is apparently capable of an exaFLOP. That means one quintillion (1018) floating-point operations per second. This makes the Dojo supercomputer one of the most powerful in the world.
The Californian automaker has worked on making this supercomputer a reality for the last couple of years or so. And if one is to believe Musk, the computer should be ready by the end of this year.
Of course, this doesn’t mean that Tesla has only been working on Dojo recently. They have developed a number of supercomputers on the journey towards developing Dojo. Andrej Karpathy, Tesla’s head of AI, has now unveiled the latest of these during a presentation at the 2021 Conference on Computer Vision and Pattern Recognition.
The Newest Supercomputer
Andrej, during his presentation, revealed the specs of this new supercomputer to the public. Tesla is claiming some really impressive features on this new cluster of supercomputers. These include:
- 720 nodes of 8x A100 80GB. (5760 GPUs total)
- 1.8 EFLOPS (720 nodes * 312 TFLOPS-FP16-A100 * 8 GPU/nodes)
- 10 PB of “hot tier” NVME storage @ 1.6 TBps
- 640 Tbps of total switching capacity
Andrej also mentioned that Tesla’s neural net architecture network has around 1.5 petabytes of data, which requires tons of computing power. For Tesla, computer vision is the bread and butter of the self-driving suite. It is what driver Autopilot and the Full Self-Driving (FSD), which is in beta mode.
The system needs to master all the data from the fleet and train the neural nets a lot. This also includes a lot of experimentation. Only then will a pure computer vision-based system work well. The current cluster is, in terms of FLOPs, roughly the fifth-most powerful supercomputer in the world, said Andrej.
He didn’t elaborate a lot on Dojo, but he did mention that it will be more powerful than this one. Musk also previously said that Tesla plans to eventually make its supercomputers available to other companies in order for them to train their neural nets on it.
If anyone had any doubts about whether Tesla is an automobile company or a technological company, this ought to clear them up. The entire point of Tesla’s vehicles is to have a medium for the company to push the most cutting-edge technology towards the public. Yes, they do make good cars and they have led the electric mobility revolution, no doubt. But it’s a vehicle built around technology, not a traditional vehicle that has technology added as a bonus.
Musk had himself mentioned recently that Tesla is more of an artificial intelligence and robotics company than an automotive company. And this supercomputer, along the Dojo to follow, nothing but proves that.