TLDR:
Elon Musk has revealed photos of the Dojo D1 Supercomputer cluster, which is equivalent to 8,000 Nvidia H100 GPUs for AI training. This supercomputer cluster uses Dojo AI accelerators and has impressive processing power for AI video training. The Dojo chips are system-on-wafer processors with a 5-by-5 array interconnected using TSMC’s InFO technology, allowing them to act as a single processor efficiently. Musk plans to have the Dojo D1 up and running by the end of the year, enhancing Tesla’s Full Self-Driving technology with the video data gathered from Tesla cars.
Article:
Elon Musk recently shared pictures of the Dojo D1 Supercomputer cluster, powered by his own Dojo AI accelerators. It is roughly equivalent to 8,000 Nvidia H100 GPUs for AI training, providing impressive processing power for AI video training. The Dojo chips, manufactured by TSMC, are system-on-wafer processors with a 5-by-5 array interconnected using InFO technology, enabling them to function as a single processor efficiently.
Musk announced during the Tesla earnings call that he would double down on Dojo development and deployment due to the high pricing of Nvidia’s GPUs. The Dojo D1 is expected to be up and running by the end of the year, complementing Tesla’s Full Self-Driving technology with the video data collected from Tesla cars.
Overall, Musk’s emphasis on AI development and deployment, particularly through the Dojo Supercomputer cluster, showcases the significant resources and effort being invested in artificial intelligence technology. The combination of different chips, including the Dojo D1, Nvidia H100, and Nvidia AI4, will bring substantial computing power to drive advancements in AI machine learning and video training.