News Technologies

Tesla in July will start building the Dojo supercomputer on its own chips – it will become one of the most powerful in the world

Tesla recently announced progress on a custom Tesla Dojo supercomputing platform built on the automaker’s own chips. Production of the supercomputer will begin in July 2023, Dojo is expected to enter the top five most advanced computing systems in the world in 2024.

Image Source: Tesla

Image Source: Tesla

Creating your own supercomputer is another important step for Tesla in the field of AI. While NVIDIA’s A100 and H100 accelerators dominate the AI ​​space at this stage, Tesla’s own AI training and inference chips could significantly reduce the company’s reliance on traditional manufacturers of such semiconductor components.

The development of the Dojo supercomputer for AI machine learning was launched at AI Day 2021. Dojo is based solely on Tesla-designed chips and infrastructure, and uses video data from Tesla’s impressive fleet of vehicles to train the neural network. The development of Tesla’s machine vision is key to autonomous driving technology. The computing power of the future supercomputer will also be used to further develop the Tesla Optimu s humanoid robot project .

The Tesla Dojo architecture uses “system-on-wafer” (System-On-Wafer), that is, the chip is a whole silicon wafer (Training Tile in Tesla terminology). Each platter holds 25 D1 accelerators and 40 I/O modules. The plate also houses the power and cooling subsystems. Tesla claims that a single system-on-a-plate replaces six GPU units and is cheaper.

While the Dojo system may not take final shape until 2024, Elon Musk is pleased with the work of his AI team, stating that Tesla’s advances in AI, both in software and hardware, go well beyond that some experts were even aware of.

See also  Exciting Sports News for 49ers Fans: Tim Ryan's Contract Extended Through the 2028 NFL Season

Software is the key to autonomous driving, and Tesla is already using a large supercomputer with NVIDIA GPUs to process data from the FSD autonomous driving system, one of the world’s most powerful supercomputing clusters.

Tesla chief engineer Tim Zaman told the public that Tesla’s compute cluster is currently 99.7% loaded, with 84% of machine time being spent on high-priority tasks. The company is in dire need of additional computing resources, and the Dojo supercomputer can dramatically improve the situation.

usiic_admin
Greetings ! I am the founder of the usiic.co platform since 2022, I am also the editor-in-chief, I hope you will find the usiic platform useful, in which I have invested a lot of time and effort. I am interested in many areas, and on the blog I share my impressions, advice and experience. I would be very grateful if you rate my post and share yours: https://usic.co
https://usiic.co/groups/news/

Leave a Reply