Tesla Devotes More Than $1 Billion to the Dojo Supercomputer – Here’s the Reason
Tesla has announced that its long-anticipated Dojo supercomputer, expected to enhance its autonomous driving efforts significantly, is set to go into production next month.
Dojo is Tesla’s proprietary supercomputer, designed from scratch for AI machine learning, specifically for video training using footage sourced from its fleet of vehicles.
While Tesla already boasts a large NVIDIA GPU-based supercomputer that ranks among the world’s most powerful, the forthcoming Dojo machine stands out with its unique chips and infrastructure that were designed in-house.
The purpose-built supercomputer is anticipated to boost Tesla’s capabilities in training neural networks using video data, a crucial component of the computer vision technology that drives its autonomous vehicles.
During Tesla’s AI Day in 2021, the firm disclosed details about its Dojo supercomputer. However, at that time, the company was still scaling up its efforts, possessing only its initial chip and training tiles, and was in the process of constructing a complete Dojo cabinet and Exapod cluster.
By AI Day 2022, Tesla unveiled some advancements with Dojo, including completing a full system tray. The company also suggested a full cluster might be ready by Q1 2023. However, the first quarter passed without announcements regarding Dojo’s operational status.
The latest update on Dojo came from Tesla’s new Tesla AI Twitter account (the fifth Twitter account initiated by Tesla since Elon Musk’s acquisition of the social media platform). The firm announced plans for Dojo to enter production in July:
Per the timeline, Tesla will continue to expand the system with more trays and cabinets until it becomes one of the world’s most powerful supercomputers by early 2024. Furthermore, it seems Tesla aims to exceed this level by the end of next year greatly.
Through Dojo, Tesla is expected to fully harness its immense database of real-world driving scenarios, compiled from millions of vehicles over several years."