Tesla has unveiled its latest version of its Dojo supercomputer, and it’s apparently so powerful it’s shut down the electrical grid in Palo Alto.
Dojo is Tesla’s own custom supercomputer platform built from the ground up for AI machine learning and specifically video training using the video data from its fleet of vehicles.
The automaker already has a large NVIDIA GPU-based supercomputer that’s among the most powerful in the world, but the new custom Dojo computer uses chips and an entire infrastructure designed by Tesla.
The custom-built supercomputer is expected to increase Tesla’s capacity to train neural networks with video data, which is critical to the computer vision technology powering its self-driving efforts.
Last year, the company unveiled its Dojo supercomputer at Tesla’s AI Day, but the company was still ramping up its efforts at the time. It had only its first chip and training tile and was still working towards building a full dojo cabinet and cluster or “exapod”.
Now, last night, during its AI Day 2022, Tesla revealed the progress made with the Dojo program over the past year.
The company confirmed that it managed to go from a chip and a tile to a system tray and a full cabinet.
Tesla claims it can replace 6 GPU boxes with a single Dojo tile, which the company says costs less than a GPU box. There are 6 of these tiles per tray.
Tesla says a single tray is the equivalent of “3 to 4 fully loaded racks of supercomputers.”
The company integrates its host interface right into the system tray to create a large complete host assembly:
Tesla can fit two of these host assembly system trays into a single Dojo cabinet.
This is what the Dojo closet looks like closed and open:
That’s where Tesla is right now, as the automaker is still developing and testing the infrastructure needed to assemble a few cabinets to build the first “Dojo Exapod.”
Bill Chang, Tesla’s principal system engineer for Dojo, said meanwhile
“We knew we had to re-examine every aspect of the data center infrastructure to support our unprecedented cooling and power density.”
They had to design their own powerful cooling and power system to power the dojo cabinets.
Chang said that Tesla tripped their local power grid’s substation while testing infrastructure earlier this year:
“Earlier this year we started load testing our power and cooling infrastructure and were able to ramp it up to over 2MW before we tripped our substation and got a call from the city.”
This is what the Tesla Dojo Exapod looks like open and closed:
Tesla released the key specs of a Dojo Exapod: 1.1 EFLOP, 1.3 TB SRAM, and 13 TB high-bandwidth DRAM.
The company used the event to try to recruit more talent, but also said it’s planning to have its full first cluster, or exapod, in Q1 2023.
It is currently planned to have 7 Dojo Exapods in Palo Alto.
Why does Tesla need a Dojo supercomputer?
It’s a valid question. Why is a car manufacturer developing the world’s most powerful supercomputer? Well, Tesla would tell you that it’s not just an automaker, it’s a technology company developing products to accelerate the transition to a sustainable economy.
Musk said it made sense to offer a dojo as a service, perhaps to rival his pal Jeff Bezos’ Amazon AWS, calling it a “service that you can use that’s available online where you can build your models much faster and.” can train for less money”.
More specifically, Tesla needs Dojo to automatically tag train videos from its fleet and train its neural networks to build its self-driving system.
Tesla realized that its approach to developing a self-driving system using neural network training on millions of videos from its customer fleet would require a lot of computing power. and it decided to create its own supercomputer to deliver that feat.
That’s the short-term goal, but Tesla will have good use for the supercomputer going forward as it has big ambitions to develop other artificial intelligence programs.
FTC: We use income earning auto affiliate links. More.
Subscribe to Electrek on YouTube for exclusive videos and subscribe to the podcast.