Tesla unveiled the latest version of its Dojo supercomputer as part of its AI Day event. The supercomputer was developed by the company’s specialists from scratch.
The Dojo is optimized for AI algorithms, in particular for machine learning using video data from Tesla electric vehicles. During testing, the power of the computer shut down the power grid in Palo Alto, reports Electrek.
The automaker already has a large supercomputer powered by Nvidia GPUs, which is one of the most powerful in the world, but Dojo’s new custom computer uses Tesla-designed chips and infrastructure.
The new supercomputer will boost Tesla’s ability to train neural networks using video data, which is critical for computer vision technology to enable self-driving driving. A year ago, the company first announced the creation of Dojo.
The upgrades to last year’s Dojo program involve going from a chip and tile system to a tray and a full cabinet. Tesla claims it can replace six GPUs with a single tile (a Dojo tile), which the company claims costs less than a single GPU. There are six of these tiles on the system tray. One pallet is equivalent to “3-4 fully loaded supercomputer racks”. Tesla can fit two of these system trays in the same chassis.
Tesla is still developing and testing the infrastructure needed to combine multiple cabinets to create the first Dojo ExaPOD, whose main specifications are already known — 1.1 EFLOP, 1.3TB SRAM, and 13TB high-bandwidth DRAM. There are plans to build seven Dojo ExaPODs in Palo Alto soon.