Tesla’s Dojo supercomputer project lead Ganesh Venkataramanan has left the company, casting some doubt over the how the project is progressing.
With Venkataramanan out, former Apple executive and current director Peter Bannon has taken over leading the project, according to a report from Bloomberg, citing sources familiar with the matter who asked to remain anonymous for discussing confidential information.
According to Venkataramanan’s LinkedIn profile, he was first hired by Tesla back in 2016, assuming the role of Director Autopilot hardware, He remained in that position for nearly three years, before advancing to his now-past position of Sr Director Autopilot Hardware, a position he held for over five years. Venkataramanan came to Tesla with a wide breadth of experience in chip making, having spent nearly 15 years at AMD.
Despite his LinkedIn profile still listing his current employer as Tesla, Venkataramanan was removed from Tesla’s internal directories this week.
Tesla’s Dojo supercomputer, first unveiled in 2020, designed to enhance neural net training for Full Self-Driving (FSD) by rapidly analyzing extensive video data from Tesla’s vehicle fleet. Additional details about this supercomputer emerged at AI Day in 2021, revealing it initially featured only one chip and training tile. The company aimed to progress towards a complete Dojo cabinet and cluster, or Exapod. AI Day 2 in the following year provided an update, indicating Tesla’s advancements and the anticipation of an operational cluster by Q1 2023.
That timeline wasn’t met, but Dojo production did start earlier this year. The company envisions Dojo becoming one of the world’s top 5 supercomputers by late January or early February 2024, projecting to reach 100 exaflops by October 2024. To put it into perspective, an exaflop signifies a supercomputer’s ability to perform at least 10^18, or one quintillion, floating-point operations per second.
Morgan Stanley predicts Tesla’s Dojo supercomputer could add $500 billion to market value
