Tesla's Supercomputer Details Revealed by Senior Director of AI Andrej Karpathy

During Tesla Inc's TSLA Autonomy Day back in April 2019, CEO Elon Musk talked of Project Dojo, Tesla's supercomputer that would be used to help train its vehicles to drive autonomously.

Tesla currently uses a combination of humans and computers to train its self-driving neural nets, and Dojo would help automate the process.

During an autonomous vehicles talk at a Computer Vision and Pattern Recognition (CVPR) conference, Tesla's Senior Director of AI Andrej Karpathy gave a talk on Tesla's current work in autonomy.

One new piece of information shared was Tesla's supercomputer is currently being used to crunch all the driving data Tesla receives from its fleet of customer vehicles currently on the road.

The specs are impressive, putting it at the 5th most powerful supercomputer worldwide, according to Electrek. The computer has 5,760 GPUs, 1.8 EFLOPS, 10 Petabytes of "hot tier" NVME storage, and 640 Terabits per second of switching capability.

In a word, it's fast. 

Despite this impressive power at Tesla's disposal, Karpathy says this is not Dojo, and he is not ready to share more information about Dojo at this time.

Tesla is using this supercomputer to aid in the training of its autonomous platform, which is now exclusively powered by cameras. 

Click here to check out Benzinga’s EV Hub for the latest electric vehicle news.

(Photo of Tesla Model S courtesy of Tesla.)