We know that Tesla is working on a supercomputer called “Dojo” which will be able to handle huge amounts of data. The goal is to develop the artificial intelligence of the Autopilot system more quickly. Dojo has not yet arrived, the debut would be expected by the end of the year, but also the current solution with which the manufacturer “trains” neural networks the basis of the Autopilot system is not far behind since according to the manufacturer it is the fifth most powerful computer in the world.
The rest, to manage the amount of data coming from the fleet of vehicles that improve assisted driving skills, a very high computing power is required. Andrej Karpathy, Tesla’s head of AI, wanted to tell something more about the supercomputer used today by his company during the Conference on Computer Vision and Pattern Recognition 2021.
YOU NEED A LOT OF CALCULATION CAPACITY
Processing the data as fast as possible is crucial for accelerate the evolution of the functionality of the Autopilot system. That’s why Tesla is investing heavily in computing power. The future “Dojo” should represent an important step forward from this point of view. An important investment, also because with the rapid increase in Tesla on the roads, the carmaker will receive an ever-increasing amount of data to process.
To get to Elon Musk’s goal of proposing in the future true autonomous driving on its electric cars, Tesla will still have a long time to work on the development of artificial intelligence. To do this, it will need ever more powerful supercomputers. Pending the arrival of Dojo, Andrej Karpathy shared some data of the latest cluster used by his company to manage and process the data with which artificial intelligence is “trained”.
- 720 nodes of 8x A100 80GB (5760 total GPUs)
- 1.8 EFLOPS (720 nodes * 312 TFLOPS-FP16-A100 * 8 GPUs / nodes)
- 10 PB di “hot tier” NVME storage @ 1,6 TBps
- 640 Tbps di switching capacity
Definitely impressive numbers. Dojo will do even better as it has been optimized for neural network development but Karpathy didn’t want to go into detail.