Tesla has been working on the Dojo supercomputer for a while and aims to make the system even more powerful. According to a new report, Tesla is doubling its Dojo D1 supercomputer chip orders with TSMC. The Dojo supercomputer is critical to Tesla’s artificial intelligence goals. Thanks to Dojo, the company aims to train and develop autonomous driving systems and artificial intelligence models internally.

Tesla’s Dojo supercomputer investment grows

The Dojo supercomputer chip, known as an application-specific integrated circuit (ASIC) developed by Tesla, is designed to train driver assistance and self-driving artificial intelligence models. Of course, the usage areas of this chip are not limited to these. It can also be used in the robotaxis targeted by Tesla. The latest report says that Tesla will double its Dojo D1 ASIC order with TSMC to 10,000 units next year. It is added that the order volume will continue to increase until 2025.

Neither Tesla nor TSMC has made a statement about the rumored Dojo D1 chip order increase. However, the report states that the Tesla supercomputer ASIC mainly uses TSMC’s 7nm family process and combines it with InFO-level system-on-wafer (SoW) advanced packaging. According to official information, the D1 ASIC is claimed to contain 50 billion transistors and is capable of 362 TeraFLOPs in FP16/CFP8 precision or approximately 22.6 TeraFLOPs in single precision FP32 tasks. This information is from a few years ago.

It can end dependence on Nvidia

However, while the construction of Doja is ongoing, Tesla is one of the largest consumers of Nvidia’s GPUs. According to 2021 data, Tesla was using 5,760 Nvidia A100 graphics processing units (GPU) for training artificial intelligence models and autonomous systems. Last month, it was reported that Tesla would purchase 10,000 new Nvidia H100 GPUs. Therefore, Dojo is important to reduce this dependence of Tesla.

One of the main features of the Dojo Project is scalability; hence Tesla may continue to order ASICs from TSMC to create multiple ‘ExaPODs’ each with 1,062,000 cores (3,000 built into the D1 chip) and capable of up to 20 ExaFLOPs. Morgan Stanley analysts suggested in their statement to CNN that Tesla will start using its artificial intelligence computing power outside of its automotive-related businesses. They predict the Dojo supercomputer will appeal to new markets and benefit the business in a similar way to what AWS did for Amazon.