Tesla’s AI training capacity is set to increase significantly by the end of 2024. The company plans to ramp up to approximately 90,000 H100 equivalent GPUs.
Tesla CEO Elon Musk gave some more details to say, “Important to note that we also use the Tesla HW4 AI computer in the training loop with Nvidia GPUs, currently at roughly a 1:2 ratio. Also, we’re changing the name from Hardware 4 (HW4) to Artificial Intelligence 4 (AI4).”
Musk also mentioned that Dojo 1, Tesla’s custom-built supercomputer, will have about 8,000 H100-equivalent GPUs online by the end of the year. He commented, “Not massive, but not trivial either,” and shared pictures of Dojo.
During Tesla’s Q2 earnings call, Musk praised Nvidia’s performance, stating, “Nvidia execution is impressive. Often hard to get Nvidia state-of-the-art GPUs.” He emphasized Tesla’s commitment to enhancing its training capabilities with Dojo, stating, “Tesla will double down with Dojo to ensure training capability. We really got to make Dojo work. And we will.”
Musk also indicated that Tesla aims to be competitive with Nvidia in AI training, underscoring the importance of making Dojo a success.
Important to note that we also use the Tesla HW4 AI computer in the training loop with Nvidia GPUs, currently at roughly a 1:2 ratio. Also, we’re changing the name from Hardware 4 (HW4) to Artificial Intelligence 4 (AI4).
That means ~90k H100, plus ~40k AI4 computers.
— Elon Musk (@elonmusk) July 23, 2024
Here’s a video shared by Musk earlier in July. This supercomputer will power AI training for Tesla for Full Self-Driving:
Tesla Gigafactory Supercomputer Cluster pic.twitter.com/aYIZ3pQ0nV
— Elon Musk (@elonmusk) July 5, 2024