Skip to content
  • by
  • News
  • 2 min read

Tesla Doubles Orders for Dojo D1 Supercomputer Chips from TSMC

Tesla is increasing its orders for the Dojo D1 supercomputer chips from Taiwan Semiconductor Manufacturing Company (TSMC), according to a report by Taiwan's Economic Daily. The custom Tesla application-specific integrated circuit (ASIC) is designed to power the Dojo supercomputer, which is used for training driver assistance and self-driving AI models. However, it is expected to have broader applications in robo-taxis and related services as well.

The report states that Tesla will double its Dojo D1 ASIC order with TSMC to 10,000 units in the coming year, with the order volume continuing to increase until 2025. The article highlights that TSMC's order momentum for high-performance computing (HPC) has been boosted by its partnership with Tesla. Although neither Tesla nor TSMC has officially confirmed the increase in chip orders, the report suggests that it is relatively common knowledge.

According to sources cited in the Economic Daily report, the Tesla supercomputer ASIC primarily uses TSMC's 7nm family process and combines it with InFO-level system-on-wafer (SoW) advanced packaging.

The Dojo D1 supercomputer chip, which was previously mentioned in an August 2021 news release, boasts impressive specifications. It reportedly contains 50 billion transistors and has a performance capability of 362 TeraFLOPs at FP16/CFP8 precision, or approximately 22.6 TeraFLOPs for single-precision FP32 tasks.

Tesla has been heavily reliant on Nvidia GPUs in the past, but the company's move towards developing its own chips showcases its desire for more customized solutions. Last year, Tesla used an unnamed Nvidia GPU cluster containing 5,760 Nvidia A100 GPUs for its Autopilot ADAS. Furthermore, Tesla is planning to activate a new supercomputer powered by 10,000 Nvidia H100 GPUs. The increased demand for Nvidia GPUs has caused supply shortages, but the development of the Dojo supercomputer is expected to help address this issue.

A key aspect of Tesla's Project Dojo is scalability, enabling the company to order more ASICs from TSMC and expand its ‘ExaPODs,' which each contain 1,062,000 cores housed in 3,000 Dojo D1 chips. These ExaPODs are capable of delivering up to 20 ExaFLOPs of computing power.

Analysts from Morgan Stanley predict that Tesla's Dojo supercomputer could open up new markets for the company beyond the automotive industry, similar to how Amazon Web Services (AWS) has benefitted Amazon. The increased AI computing power provided by Dojo is expected to enhance Tesla's business prospects.

Sources:

– Taiwan's Economic Daily
– CNN
(Note: URLs have been removed)