Tesla Shows Off its Unique New Generation Supercomputer

Tesla’s unique on-board new supercomputer to train vehicles autopilot features.



Tesla just launched the world’s fifth most powerful computer- ‘the supercomputer’ that is created by Tesla’s third supercomputer cluster. This supercomputer is way ahead of the expected Dojo supercomputer presentation which is responsible for neural network Artificial Intelligence training. It is the paradigm of Tesla’s upcoming Dojo Project. Awaiting patiently for the Dojo super computer launch, the company astonishingly divulged another quantum leap in the field of technology. The supercomputer is a part of the promotion by Musk for a vision-only approach. Tesla’s supercomputer is introduced for developing autonomous driving which is the company’s upcoming self-driving Artificial Intelligence.

Tesla has been operating on its auto self-driving program for many years and during all this time it has directed and trained its autopilot and the whole auto self-driving suit with the help of Neural Network to style it as more exact and vigorous in its movement. This led to the establishment of Tesla as one of the companies with the strongest semi-autonomous driving programs across the globe.

On 21st June at the 2021 Conference on Computer Vision and Pattern Recognition, the head of Artificial Intelligence of Tesla, Andrej Karpathy divulged the new supercomputer that enables the car manufacturer to discard the tracking system and lidar sensors in favor of high-quality optical cameras on self-driving cars because cameras are faster than radar or lidar. To get a computer to reciprocate to a new environment appropriately so that the humans can access and a gigantically powerful supercomputer to direct the company’s autonomous driving technology using that data set explained Karpathy.

The supercomputer launched by Tesla is a high-performance, GPU accelerated computing. It is for the transformation in HPC (High-Performance Computing) at the STFC Rutherford Appleton Laboratory in Didcot. ‘Emerald’ was the name designated to the 84-node cluster and it uses accelerators which is of 372 Nvidia Tesla M2090.

Tesla’s newest generation supercomputer has 640 TBPS of total switching capacity, 720 nodes of 8x A100 80GB. (5760 GPUs total). According to Karpathy it has 10 petabytes of “hot tier” NVME storage and runs at 1.6 TBPS per second. With 1.8 EFLOPS (720 nodes * 312 TFLOPS-FP16-A100 * 8 GPU/nodes) it might be the fifth most powerful supercomputer in the world.

Tesla’s vehicles have already proven a feature that is called pedal misapplication mitigation, where the car recognizes pedestrians and amblers in its path and possibly saving the pedestrians in front of the vehicle and also protecting the driver from accelerating into a river. The supercomputer of Tesla collects video from around 8 cameras surrounding the vehicle at 36 frames per second. This provides a huge amount of information about the surroundings of the car.

Another Tesla Artificial Intelligence team’s innovation has been auto-labeling. With this auto-labeling technique it can robotically label things like objects, road hazards and this is done with the help of the videos captured by vehicles on a Tesla Camera, unlike the time-consuming large data sets that require manual labeling. Now with the supercomputer, Tesla has assembled one million videos which are of around 10 seconds each and has labeled 6 billion objects along with depth, velocity, and acceleration. Hence all this takes up monstrous storage of 1.5 petabytes.


Must see news