Tesla Starts to Produce Dojo Supercomputer to Train Driverless Cars

In its second quarter earnings report for 2023, Tesla has announced the commencement of production for its Dojo supercomputer, a critical component for training its fleet of autonomous vehicles. The company outlined four main technology pillars necessary to achieve vehicle autonomy at scale: an extensive real-world dataset, neural net training, vehicle hardware, and vehicle software, and emphasized that they are internally developing each of these pillars.

‍

The Dojo training computer represents a significant advancement towards faster and more cost-effective neural net training. While Tesla already possesses a powerful Nvidia GPU-based supercomputer, the Dojo custom-built computer utilizes chips specifically designed by Tesla. CEO Elon Musk had previously dubbed this high-performance training computer as "Dojo," envisioning it to be capable of an exaflop, equivalent to 1 quintillion (10^18) floating-point operations per second. This immense computing power is difficult to fathom, as it would take over 31 billion years to perform the same number of calculations sequentially.

‍

Tesla's progress on the Dojo project was shared during its AI Day in 2021, where executives unveiled its initial chip and training tiles, which were slated to form a complete Dojo cluster or "exapod." The plan involved combining two sets of three tiles in a tray and placing two trays in a computer cabinet, resulting in over 100 petaflops per cabinet. With a 10-cabinet system, Tesla aimed to break the exaflop compute barrier with its Dojo exapod.

‍

Subsequent updates at AI Day 2022 showcased further advancements, including the presentation of a full system tray for the Dojo. Tesla had previously projected a full cluster by early 2023, though the latest reports indicate that the completion of the Dojo exapod will likely occur in early 2024.

‍

Tesla's commitment to developing its Dojo supercomputer demonstrates its dedication to pushing the boundaries of autonomous vehicle technology through innovative in-house solutions. By leveraging this powerful tool, the company aims to make significant strides in training its fleet of autonomous vehicles, ultimately driving advancements in the field of vehicle autonomy at scale.

‍