Tesla is set to unveil its highly anticipated supercomputer, powered by Nvidia H100, on Monday, according to a report by @sawyermerRritt. The new supercomputer will be utilized for a variety of artificial intelligence (AI) tasks and is also capable of solving high-performance computing problems. With a staggering 10,000 NVIDIA H100 video cards, the supercomputer’s peak performance will reach an impressive 340 FP64 PFLOPS, surpassing even the Leonardo supercomputer, currently ranked fourth in the world for productivity.
Tesla’s primary objective with this supercomputer is to accelerate the training of its full self-driving (FSD) technology. “By training on real video data, we may have the largest dataset for learning in the world,” said Tim Zaman, Tesla’s AI Engineering Work Manager.
However, Tesla is facing supply difficulties due to the high demand for the NVIDIA H100 graphics cards. To combat this issue, the company has invested over $1 billion in the development of its own supercomputer dubbed “Dojo.” Besides facilitating FSD training, Dojo will also process data from Tesla’s entire fleet of vehicles.
Furthermore, Elon Musk recently announced Tesla’s plans to invest over $2 billion in AI training in 2023, followed by another $2 billion in 2024 for FSD training computations. This highlights Tesla’s determination to overcome computational challenges and gain a significant competitive edge.