Tesla releases self-developed AI chip D1

On the ongoing Tesla AI Day, Tesla released a custom chip D1 for training artificial intelligence networks in data centers. Tesla Director Ganesh Venkataramanan said at the Artificial Intelligence Day event that Tesla CEO Elon Musk asked Tesla engineers to design a super-high-speed training computer a few years ago. This is how we started the Dojo project.

The D1 chip is part of Tesla’s Dojo supercomputer system. It uses a 7-nanometer manufacturing process and has a processing capacity of 1024 Gigabits. Venkataramanan said that Tesla puts a group of such chips on a training tile to provide 9 petaflops of computing power. 120 tiles are placed together in several server cabinets, which is equivalent to more than one exaflops.

These chips can help train models to identify various objects from the video data collected by the internal cameras of Tesla cars. Model training requires a lot of computational work. Musk said that Tesla should be able to run the artificial intelligence training computer DOJO chip next year. It is open to licensing artificial intelligence technology to other automakers, and it is unlikely to open-source Tesla’s AI chips.

Leave a Comment