Tesla’s new AI supercomputer
The Tesla AI supercomputer, which features 10,000 Nvidia H100 compute GPUs and is valued at over $300 million, is set to go live at the automaker today, and will enhance its computing capabilities for training its full self-driving (FSD) technology. With a peak performance of 340 FP64 PFLOPS for technical computing, and 39.58 INT8 ExaFLOPS for AI, the cluster will be one of the world’s most powerful machines, and will enable Tesla to handle large training datasets and have a cache capacity beyond 200PB.
This comes as no surprise, as Nvidia has long been the go-to provider of GPUs for deep learning and AI, and Tesla has been an early adopter and staunch supporter of the company’s hardware. However, the demand for these GPUs has been so high that Nvidia is struggling to meet it, prompting Tesla to invest in developing its own supercomputer called Dojo, which will accelerate FSD training and manage data processing for Tesla’s entire vehicle fleet.
“With the new Tesla AI HPC Cluster going live, Tesla’s computing capabilities will be stronger than ever,” says Tesla. “The cluster will enable Tesla to train faster and manage larger training datasets.”
Bringing both the Nvidia H100 GPU cluster and Dojo online simultaneously will give the company unmatched computing power in the automotive industry, and will enable Tesla to gain an edge over its competitors.
Elon Musk has revealed plans to spend over $2 billion on AI training and computing for FSD training in 2023 and 2024, which demonstrates Tesla’s commitment to overcoming computational bottlenecks and gaining an edge over its competitors.
You can check out the full details of Tesla’s new AI supercomputer here.