Tesla announced in its second quarter earnings report for 2023 that it has begun production of its Dojo supercomputer, which will be used to train its fleet of autonomous vehicles. The company identified four main technology pillars needed to achieve vehicle autonomy at scale: extremely large real-world datasets, neural net training, vehicle hardware, and vehicle software.
Tesla stated that it is developing each of these pillars in-house, and the production of the Dojo training computer marks a significant step towards faster and cheaper neural net training.
The Dojo supercomputer is a custom-built computer that utilizes chips designed by Tesla. While the automaker already boasts one of the most powerful Nvidia GPU-based supercomputers in the world, the Dojo represents a new level of performance and capability. Elon Musk, Tesla’s CEO, named the Dojo supercomputer in 2019 and claimed that it would be capable of an exaflop, or 1 quintillion floating-point operations per second.
To put this into perspective, a one exaflop system could perform calculations at a rate that would take over 31 billion years to match if done sequentially. This level of computational power is essential for training the neural networks used in Tesla’s autonomous driving technology.
Related Post
At Tesla’s AI Day in 2021, Dojo was still in development. The company revealed its first chip and training tiles, laying the foundation for a full Dojo cluster or “ExaPod.” Tesla plans to combine multiple tiles in a tray and multiple trays in a computer cabinet to achieve over 100 PFlops (petaflops) per cabinet. With a 10-cabinet system, Tesla’s Dojo ExaPod will break the exaflop barrier.
By AI Day 2022, Tesla had made progress on Dojo and had a full system tray in place. The company projected that a full cluster would be available by early 2023, but it now appears that it may be early 2024 by the time the Dojo ExaPod is fully operational.
The Dojo supercomputer represents a significant investment by Tesla in advancing its autonomous driving technology. With access to large real-world datasets, Tesla’s neural networks can continue to improve and refine their ability to accurately perceive and navigate the world.
The Dojo’s computational power enables Tesla to run complex simulations and training scenarios, accelerating the development of its vehicle hardware and software. This can lead to significant advancements in the safety and capabilities of Tesla’s Autopilot and Full Self-Driving features.
Training autonomous vehicles to be capable of safely navigating various real-world scenarios requires immense computational resources. Tesla’s in-house development of the Dojo supercomputer shows the company’s commitment to pushing the boundaries of what is possible in autonomous driving technology.
As the production of the Dojo supercomputer ramps up, Tesla will have access to even greater compute power, allowing for more advanced neural network training and accelerated progress towards achieving full vehicle autonomy.
The Dojo project is a testament to Tesla’s ambitious goals and its determination to stay at the forefront of technological innovation. By developing its own supercomputer, Tesla can tailor the hardware and software to its specific needs, optimizing performance and efficiency.
Ultimately, the Dojo supercomputer represents a significant milestone in Tesla’s journey towards fully autonomous vehicles. With its impressive computing power, Tesla aims to train its neural networks more quickly and effectively, enabling its vehicles to operate autonomously in diverse and complex environments.
The successful deployment of the Dojo supercomputer will have far-reaching implications not only for Tesla but also for the future of autonomous driving. By pushing the boundaries of compute capabilities, Tesla is driving the industry forward and shaping the future of transportation.
