Tesla is powering up its $300 Million AI Supercomputer Today

Tesla's making a significant power move today as it prepares to bring its brand-new new AI-cluster online. Rocking a hefty 10,000 Nvidia H100 compute GPUs, the machine will tackle high-performance computing (HPC) workloads and AI applications, placing Tesla's capabilities among the global AI elite.

If you want to stay on top of the latest trends and insights in AI and tech, look here first.

https://preview.redd.it/c5ykgmr6r6lb1.png?width=970&format=png&auto=webp&s=8b830c8754c1a11792149f57d19a57a77fc8b161

Here’s why this matters:

• This Nvidia H100-based AI supercomputer will be one of the most powerful globally. With a peak performance of 340 FP64 PFLOPS and 39.58 INT8 ExaFLOPS for AI programs, even Leonardo, currently the fourth highest-performing supercomputer, is surpassed.
• Tesla’s full self-driving (FSD) technology training pace will get a major boost. Given the accelerating race for autonomous vehicles, this strategic move could amplify Tesla’s competitive edge and establish the company as the possessor of one of the world's fastest supercomputers.
• Tesla’s subsequent project, Dojo, is also making headlines as the company invests $1 billion to develop the supercomputer on its own. The combination of the new Nvidia GPU cluster and Dojo will offer unmatched computing power in the automotive space.

As per AI Infra & AI Platform Engineering Manager Tim Zaman at Tesla:

• Tesla potentially owns the world's largest training datasets thanks to real-world video training, with hot tier cache capacity exceeding any other LLM's by a considerable margin.

Long-term directions:

• Tesla’s commitment to advancing in AI and computation is unrelenting. Elon Musk recently disclosed plans to spend over $4 billion over the next two years on AI training and computing for FSD training, demonstrating Tesla’s resolve to overcome computational bottlenecks and gain significant advantages over rivals.

P.S. If you like this kind of analysis, I write a free newsletter that tracks the most relevant news and research in AI and tech---stay updated in under 3 mins/day.

(source)