Tesla Investing $500M in Buffalo Factory’s Dojo Supercomputer Project

Tesla announced on January 26, 2024, that it is now making plans to build a giant Dojo supercomputer at its Buffalo Gigagfactory in New York. The company has confirmed that it will spend $500 million on this project as it will also give opportunity to 1,500 people to find jobs. Moreover, Tesla has revealed that this project will be able to generate 1.5 exaflops of computing power. Based on these figures, it is expected that this move by Tesla will result in the development of one of the most powerful supercomputers in the world.

Basis of Announcement

Kathy Hochul, the State Governor of New York, announced the Dojo supercomputer project right after the CEO of Tesla, Elon Musk, revealed the primary details of the Dojo architecture at the Tesla AI Day event. The CEO of Tesla has clarified that this supercomputer is designed specifically to process video data (expected to be in millions of terabytes) from its more than 4 million cars and stored in real-life driving stations.

In response to the Governor’s statement, Elon Musk went to Twitter and confirmed the announcement. He stated in a post that the Governor hasn’t made any false statements. Musk said that the investment of $500 million is a large sum and is equivalent to a 10k H100 Nvidia system. Additionally, Tesla’s CEO has revealed that the company is looking forward to spending more on Nvidia hardware to stay competitive in the world of AI. As per Musk’s post details, the table stakes have touched the mark of several billion dollars so far.

Tesla Dojo Supercomputer Project

Tesla Dojo is generally a recognition and computer vision video processing supercomputer designed by Tesla. The primary function of this computer is to train the company’s learning models to improve the driver-assistance systems of its fully self-driving cars. The company has revealed that the project went under construction back in July 2023. However, Musk is looking forward to expanding the functioning of this Dojo supercomputer to its Buffalo Gigagfactory in New York.

The architecture of the Dojo computer is also different from that of other traditional supercomputers. It integrates numerous custom chips called D1. Each of them has 354 billion transistors and 1.25 terabytes of SRAM. These chips interconnect with each other to ensure smooth and fast communication to share the data. Hence, the requirement for virtual memory, coherency, and global lookup directories will be diminished.

Besides, every D1 chip inside this Dojo supercomputer will have 362 teraOPS of computation. A total of 25 chips will make the DPU or Dojo Processing Unit and is capable of performing 9 petaops. Based on the number of DPUs combined with each other, different components are formed which are able to deliver different functionalities.

  • Dojo Training Tile (DTT): 10 DPUs (90 petaOPS)
  • Dojo Training Matrix (DTM): 120 DTTs (1.8 exaOPS)

On combining a total of 140 DTMs, Tesla has made a Dojo supercomputer that can perform 1.5 exaflops.

Future Prospects

The launching of this Dojo supercomputer will train the machine learning models of the full self-driving systems of Tesla to enhance the functionality of its driverless cars. Tesla is also claimining that the trained models will deliver higher accuracy and ensure lower latency to handle complex tasks, including the likes of 4D video reconstruction, simulation, and multi-agent planning, smoothly and effectively.

Michael Clark

Michael Clark has been a ghostwriter for 5 years. Expert in tech trends, SEO & business marketing-related content. He has always wanted to pursue writing as a career. Michael has written many articles, eBooks, blogs, and other content for many websites across different industries. He is highly experienced in SEO, article marketing, and website content writing.

Related Articles

Back to top button