Skip to content
Home » Tesla AI Power Move at CVPR : Plans to Achieve 100 Exa-Flops by 2024

Tesla AI Power Move at CVPR : Plans to Achieve 100 Exa-Flops by 2024

tesla

Tesla AI is making waves again, this time with their recent announcement at the CVPR conference. Ashok’s PowerPoint presentation unveiled Tesla’s ambitious plan to achieve a total computing power of 100 Exa-Flops by the end of 2024. This is a significant increase from their current equivalent of 14,000 Nvidia A100s, but with the launch of Dojo next month, their computing power is expected to grow rapidly.

By February 2024, Tesla’s computing power is expected to enter the top five in the world, and by the end of that year, they plan to reach their goal of 100 Exa-Flops. This is equivalent to 300,000 Nvidia A100s and represents a tenfold increase in computing power.

Tim Zaman, the head of Tesla’s training infrastructure, revealed that the current idle time of Tesla’s training cluster is only 0.3%. With 100 Exa-Flops, it is expected that Tesla can cover the computing needs of FSD Beta in Asia Pacific and Europe, as well as for Tesla Bot.

At the CVPR conference, Tesla showcased the neural networks and tech behind Autopilot, Project Optimus (AKA the Tesla Bot), and their Dojo supercomputer. These multi-modal neural nets can take in data from cameras, maps, navigation systems, motion sensors and GPS to predict things like how occupied the roads are and collision risks. This tech is already cruising around in Tesla vehicles today!

How’s Tesla pulling this off? At an insane scale, using footage from multiple trips on the same route to fully reconstruct environments. This lets Tesla predict possible scenarios based on what’s happened at that location before. Consistently, across all their camera angles.

All of this is made possible by fleet-scale auto-labeling, where video data from multiple trips in the same location can be used to reconstruct the entire scene. This enables Tesla to predict possible outcomes given past observations in a consistent manner across multiple camera views.

Tesla is not just designing AI models, but they’re also integrating them with hardware, just like how Apple did with their computers. With Tesla’s AI talent and engineering team, this integration is expected to make a significant impact.

Tesla’s latest revelations prove they’ve got lofty aspirations far beyond electric cars.

Related Posts:
Tesla FSD Beta 11.4.4 is now available for download!

Tesla Bot’s Latest Stride: Teaching Itself How to Grab Stuff Better