Dojo

Why does Tesla want to build its own $1 billion supercomputer? https://qz.com/elon-musk-tesla-supercomputer-evs-data-project-dojo-1850658578 Tesla's Dojo supercomputer will handle petabytes of data generated by its cars' Autopilot and Full Self-Driving modes

—

Morgan Stanley: Tesla's Dojo "team has a combined 250+ years of relevant hardware & software experience between 12 of the notable engineers working on Dojo" https://twitter.com/SawyerMerritt/status/1701041330355347729

—

There seems to be some confusion about Tesla's new 10,000 unit H100 GPU cluster that they just turned on yesterday, and their own in-house Dojo supercomputer that started production last month.

These are two separate things.

Consider Tesla’s NVIDIA H100 GPU cluster like Apple using Intel chips. They are powerful, but not as optimized as Tesla would like. Now, consider Tesla’s own vertically integrated Dojo supercomputer/system like Apple’s in-house designed Silicon chips, which are hyper-optimized and specifically designed to bring next-level performance and efficiency. This will enable Tesla to improve FSD faster and unlock next-level compute capabilities.

But Tesla will use both to improve FSD.

By next year, Tesla's Dojo supercomputer is projected to be one of the most powerful supercomputers in the world. https://twitter.com/SawyerMerritt/status/1696382633398186454