Tesla will begin production of its stackable Dojo supercomputer next month, the company announced in a Twitter thread from a new Tesla AI account.
Dojo, which was first announced during Tesla’s AI day in 2021, is a fully scalable in-house designed and built supercomputer. Dojo will help accelerate the development of full self-driving by enabling Tesla to scale up the number of real-world driving videos to train Tesla’s FSD neural net software.
Dojo has been specifically designed for AI machine learning and for training using video data coming from Tesla’s 3 million plus EVs.
And will be trained on enormous amounts of compute pic.twitter.com/BsmG9Vse6I
— Tesla AI (@Tesla_AI) June 21, 2023
The Twitter thread contains a graph showing the total amount of compute Tesla has available to train its AI autonomous driving software.
With the production of the new scalable computer, the amount of compute available to Tesla is now set to grow dramatically in coming years. By early 2024 Tesla predicts Dojo will be one of the top 5 supercomputers in the world.
Until now, Tesla has been using a large NVIDIA GPU-based supercomputer which is also one of the most powerful in the world.
The imminent production of Dojo as a scaleable supercomputer designed specifically for neural net training could be a game changer for the future of Tesla FSD and for AI more broadly.
Although developed for driving applications, Tesla’s FSD software is also being utilised in Tesla’s humanoid robot. @Tesla_AI posted a video of the humanoid navigating its environment using the software.
Tesla is building the foundation models for autonomous robots pic.twitter.com/VUES9jU3ze
— Tesla AI (@Tesla_AI) June 21, 2023
Daniel Bleakley is a clean technology researcher and advocate with a background in engineering and business. He has a strong interest in electric vehicles, renewable energy, manufacturing and public policy.