Tesla shares update and new details on Dojo supercomputer

Tesla has shared an update and new details on its Dojo supercomputer, saying that it will go into production in July. The update came from Tesla’s new artificial intelligence Twitter account @TeslaAI, with CEO Elon Musk later adding more information to say Dojo has been up and running for a few months.

Musk first revealed the existence of the Dojo supercomputer in 2020, saying it will be used to optimize neural net training both for Full Self-Driving (FSD) by being able to quickly analyze the extensive video data from Tesla’s fleet of vehicles. We then got more information of this state of the art computer a year later at AI Day in 2021, learning at that time it only had one chip and training tile and the company was still working towards a full Dojo cabinet and cluster, or Exapod. The next update came at AI Day 2 last year with Tesla revealing it had made progress and was on track to have an operational cluster by Q1 2023.

That timeline wasn’t quite met, but it was close. According to the Tesla AI Twitter account, the start of Dojo production will be next month, with the company continuing to add to it and projecting it will become one of the top 5 supercomputers in the world by late January or early February 2024. But the company isn’t planning on stopping there, with the graph showing they plan to reach 100 exaflops by October 2024. If you’re wondering what an exaflop is, you’re not alone – one exaflop is when a supercomputer can calculate at least 10¹⁸ or one quintillion floating point operations per second.

And will be trained on enormous amounts of compute pic.twitter.com/BsmG9Vse6I

— Tesla AI (@Tesla_AI) June 21, 2023

While the official start will be next month, Musk revealed the supercomputer has been up and running “for a few months.” However it has only been running “useful tasks” during that time, and the not the full deep neural net training it was primarily designed for.

With Dojo starting next month, it will begin processing huge amounts of video data from Tesla’s fleet of vehicles to improve the self-driving software, although it is unclear how quickly we will see any improvements in the behaviours of FSD Beta.

Click on the tweet below to see the full Tesla AI Twitter thread.