Elon Musk Unveils $300M Tesla Supercomputer With 10K Nvidia Chips
Elon Musk has powered up a new supercomputer for Tesla that could help it train cars to drive autonomously. Musk hinted at the plan in a video last weekend on X (formerly Twitter) before Tesla AI platform engineering manager Tim Zaman confirmed it. The Supercomputer includes more than 10,000 Nvidia H100 graphics processing units (GPUs) for 340 petaflops of performance and nearly 40 exaflops of AI throughput.
— Elon Musk (@elonmusk) August 26, 2023
As the video above shows, Tesla intends to leverage the extreme power of the supercomputer for training its self-driving software based on its vast volumes of real-world driving data. The dramatic increase in speed and throughput promises to give Tesla key competitive advantages in autonomous vehicle development over rivals. Speculation about Musk’s interest in generative AI has focused on the renamed Twitter and the new X AI firm he launched in July. The news reveals the shape of Musk’s plans for the Nvidia GPUs that seem ideal for large language model training. GPUs are both crucial and expensive, with a price of at least $10,000 each. The new cluster exemplifies the extreme resources required at the cutting edge of AI currently. But by minimizing training lags, Tesla aims to accelerate the development and commercialization of full self-driving capabilities.
However, with strong demand for the H100 GPUs, Tesla is also investing over $1 billion in custom chips powering its Dojo supercomputer. Dojo will work alongside the new Nvidia-based system to manage data processing for Tesla’s entire fleet. The combined supercomputing muscle underscores Tesla’s determination to overcome barriers limiting AI progress. CEO Elon Musk stated the company will spend $4 billion on computing for AI training in 2023-2024 alone. With autonomy viewed as critical to Tesla’s future, it is pursuing an unmatched computing scale.
Of course, Musk’s pursuit of generative AI stands as a striking contrast to his setup and signing of an open letter calling for a months-long pause on generative AI training. Hundreds of people, including celebrities and tech pioneers like Steve Wozniak, all signed the request. He’s also frequently criticized OpenAI and its ChatGPT generative AI chatbot, despite his history as a founder of OpenAI who donated $100 million to the company when it was a non-profit. He left in 2018 and has been outspoken against its change to a for-profit model and its decision to partner with Microsoft.