In an industry-leading performance of 1.8 exaflops, Andrej Karpathy (Senior director of AI at Tesla) unveiled the Tesla in-house supercomputer this week at CVPR that is being used to train deep neural networks for Autopilot and self-driving capabilities using 720 nodes powered by NVIDIA A100 Tensor Core GPUs (5,760 total).
Tesla is making it possible for engineers to do their life’s work efficiently and at the cutting edge by providing unprecedented levels of computing.
With NVIDIA Ampere, the A100 GPU delivers up to 20x higher performance than its predecessor and can be partitioned into seven GPU instances. This means that no matter what your data center needs are today, they will never have a problem with this product!
The GPU cluster is part of Tesla’s integrated approach to autonomous driving, which uses the 1 million cars already on the road to refine and build new features.
The Tesla has a deep neural network running in the car that perceives and makes predictions while you’re driving. Of course, it does this without actually controlling your vehicle, but it still manages to do so much for you. Tesla engineers refine their DNN by training it to identify difficult cases and allowing the system to compare its predictions with reality.
When all 1 million 10-second clips have been recorded, the DNN is then run through these scenarios repeatedly in a data center until it operates without making mistakes. Then, finally, it’s sent back to the vehicle where we can begin driving again.
Tesla has invested heavily in making its cars autonomous, not just for safety but also to help engineers experiment and iterate more quickly.
According to Karpathy (Senior director of AI at Tesla), the current DNN structure allows a team of 20 engineers to work on one network at once, which separates different features for elementary development.
Workshop on Autonomous Driving at CVPR’21: https://www.youtube.com/watch?v=eOL_rCK59ZI&t=29533s
Source: https://blogs.nvidia.com/blog/2021/06/22/tesla-av-training-supercomputer-nvidia-a100-gpus/
Asif Razzaq is the CEO of Marktechpost, LLC. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. His most recent endeavor is the launch of an Artificial Intelligence Media Platform, Marktechpost, which stands out for its in-depth coverage of machine learning and deep learning news that is both technically sound and easily understandable by a wide audience. The platform boasts of over a million monthly views, illustrating its popularity among audiences.
