Tesla claims to have begun manufacturing the Dojo supercomputer that will be used to train its fleet of autonomous vehicles.
In its second quarter earnings report for 2023, the business defined four main technology pillars required to solve vehicle autonomy at scale: neural net training, vehicle hardware, extremely large real-world dataset, and vehicle software.
The company stated in its report, “We are internally building each of these pillars. With the beginning of Dojo training computer production, we are moving towards quicker and less expensive neural network training.
The automaker already owns a sizable supercomputer powered by Nvidia GPUs that is among the most potent in the world, but the new Dojo custom-built computer uses chips designed by Tesla. Elon Musk, the CEO of Tesla, gave this “super powerful training computer” the name “Dojo” in 2019.
In the past, Musk has asserted that Dojo will be able to process 1 quintillion (1018) floating-point operations per second, or an exaflop. That much power is astonishing. According to Network World, one would have to conduct one calculation per second for 31,688,765,000 years to match what a one exaFLOP computer system can do in just one second.
Tesla revealed Dojo’s advancements at AI Day 2022, including the addition of a full system tray. The manufacturer had previously stated that a complete cluster will be available by early 2023, but it now appears that this will most likely happen in early 2024.