Sep 01, 2023, 02:35 PM
This week marked the launch of the Dojo supercomputer, which has been assigned the formidable task of processing the extensive volume of data gathered by Tesla's fleet of vehicles. Through this ambitious venture, the automaker aspires to make significant strides toward realizing the vision of a fully autonomous car. While Tesla has been diligently pursuing autonomous driving for years, it is important to note that their current devices do not yet achieve complete autonomy. Back in July, CEO Elon Musk announced a substantial investment of $1 billion for the development of the Dojo computer from the ground up. This supercomputer has undergone extensive development and features Tesla's proprietary D1 chips, complemented by various Nvidia AI chips like the A100, which have been incorporated over time. What sets the Dojo supercomputer apart is its remarkable computational prowess, comprising a staggering 10,000 Nvidia H100 chips. These H100 chips are in high demand for fueling substantial machine learning projects. Valued at over $300 million, the AI cluster possesses computational capabilities that rival the most potent supercomputers currently utilized by public universities and other institutions. It's noteworthy that the entire cluster, managed by Tim Zaman, responsible for overseeing Tesla's AI infrastructure, is situated on-site within Tesla's facilities and is not hosted in the cloud.
