NVIDIA has unveiled the DGX SuperPOD, the world’s 22^nd fastest supercomputer designed to provide AI infrastructure for autonomous vehicles. Based on 96 NVIDIA DGX-2H supercomputers and Mellanox interconnect technology, the DGX SuperPOD provides 9.4 petaflops of computing horsepower.

The NVIDIA DGX SuperPOD is powered by 1,536 NVIDIA V100 Tensor Core GPUs connected by the NVIDIA NVSwitch and Mellanox network fabric. The system can train a ResNet-50 model in less than two minutes, according to NVIDIA, 18,000x faster than solutions of just four years ago. The company also states that the DGX-2H is 400x smaller than other TOP500 supercomputers.

The system was built in three weeks, NVIDIA says. The company plans to apply it to autonomous vehicle use cases that require neural networks be retrained tens of thousands of times to deliver the extreme accuracy needed in automotive systems.

Organizations looking to take advantage of the NVIDIA SuperPOD architecture can access it through the DGX-Ready Data Center program.