Tesla Demonstrates Cybertruck-Inspired Server Racks for Self-Driving Software Training
In a significant development, Tesla, the leading electric vehicle manufacturer, has announced a shift in its focus away from the Dojo supercomputer project, which was aimed at creating an in-house AI training supercomputer. The decision comes as the company is set to deploy a 300,000 B200 GPU supercomputer by summer 2025, according to xAI, indicating a further purchase of 200,000 B200 GPUs.
The design and setup of the server cabinets, unveiled with a quote from Tesla AI: "Quiet please. The machines are learning," bear a striking resemblance to the pictures of the server cabinets Tesla shared last year of its Dojo 1 supercomputer. However, it's likely that these units are no longer a priority given the overall shift away from Dojo.
Tesla is increasing its reliance on external tech partners like Nvidia, AMD, and Samsung for compute and manufacturing needs, signaling a move away from in-house chip development for high-performance AI training. The server cabinets, part of Tesla's in-house designed hardware for AI training, are said to contain these GPUs, but it's possible they may also contain next-gen D2 chips, which Tesla had planned for the Dojo project before its cancellation.
Tesla is not the only one investing in powerful AI hardware. xAI, another company in the AI sphere, is using 100,000 H100 GPUs in its Memphis Supercluster as of July 2024. Meanwhile, Tesla is operating 35,000 Nvidia H100 GPUs as of April 2024, with plans to increase this to 85,000 by the end of 2024. The company also plans to purchase an additional 50,000 H100 GPUs, with 50,000 allocated for Giga Texas to support Full Self-Driving (FSD) training.
The design and setup of the Dojo supercomputer are likely to contribute significantly to the advancement of Tesla's autonomous driving technology. The clean wiring in the server cabinets, described as next-level, and the super sleek server cabinets with a Cyber design, inspired by the Cybertruck, are testament to Tesla's commitment to innovation and cutting-edge technology.
While the spirit of Dojo might live on through the use of next-generation vehicle chips, the original project focusing on the Dojo supercomputer and its associated chips is no longer a priority. The current status of Tesla's Dojo supercomputer setup is that it has been effectively shut down. Elon Musk has confirmed that Tesla is ending its Dojo project, opting instead to focus on using Tesla's AI5 and AI6 chips for training purposes.
In terms of cost, Tesla's 2024 Nvidia spending is estimated to be between $3-4 billion. This investment in powerful AI hardware is a testament to Tesla's commitment to pushing the boundaries of what is possible in the realm of autonomous driving technology.
[1] Tesla ends Dojo project, focuses on AI5 and AI6 chips for training [2] Tesla's Dojo supercomputer: What we know so far [3] Tesla's AI6 chip: What we know so far [4] Tesla's AI chip strategy: What we know so far [5] Elon Musk hints at the future of Dojo in Tesla's AI chips
- Although Tesla has ended its Dojo supercomputer project, the company is still heavily investing in artificial-intelligence, with plans to use its in-house AI5 and AI6 chips for training purposes and to purchase additional 50,000 Nvidia H100 GPUs to support Full Self-Driving (FSD) training.
- The design and setup of the server cabinets containing Tesla's in-house AI training hardware, which bear a resemblance to the server cabinets of the Dojo 1 supercomputer, showcase Tesla's ongoing commitment to technology and cutting-edge artificial-intelligence.
