This month the “Computer vision and pattern recognition” conference was held and Tesla showed the supercomputer used to train the AI that will control Tesla’s self-driving cars.
During the conference, Andrej Karpathy, head of AI development at Tesla, said that their supercomputer is the fifth most powerful in the world. It has 5,760 graphics processors (GPUs) with 1.8 exaflops computing power and 10 petabytes of NVMe storage with a maximum transfer rate of 1.6 terabytes per second.
During the conference, Karpathy also emphasized that they do not use LIDAR sensors for video in their self-driving system. Tesla believes that LIDAR, which stands for Light Detection And Ranging, is not scalable at the level they need to be.
“We take a video-based approach, which of course is much more difficult because you have to have neural networks that work extremely well with video, but when that is in place you have a general system that works with video and it can be used anywhere on the planet” Karpathy said during the conference.
The supercomputer that was unveiled during the conference is the forerunner of the even sharper computer “Dojo” that Tesla is currently developing. During the conference, Karpathy said:
“Next up is Dojo, which we are currently working on, but I am not ready to reveal anything more right now.”
Already in 2019, the plans for the Dojo were revealed and in 2020, Elon Musk tweeted about the project.
Tesla is developing a NN training computer called Dojo to process truly vast amounts of video data. It’s a beast! Please consider joining our AI or computer/chip teams if this sounds interesting.
— Elon Musk (@elonmusk) August 14, 2020
