Nvidia Touts Supercomputer for Autonomous Vehicle Development

Nvidia Corp. says its new DGX SuperPOD processing system’s 9.4 petaflops of processing power can speed the development of self-driving car systems.

Nvidia notes that a single self-driving car could generate 1 terabyte of data per hour.

SuperPOD aggregates the power of 96 Nvidia DGX-2H supercomputers with more than 1,500 interconnected graphic processing units. The network took just three weeks to assemble and is only 0.2% the size of rival systems.

Using artificial intelligence and neural networks, the DGX system can be used to update sophisticated software algorithms, find potential failures and retrain how self-driving cars react to different driving scenarios based on real-time inputs.

Nvidia says the new system is more than 18,000 times faster than the company’s former state-of-the-art single K80 GPU. What took that unit 25 days to compute can be done in less than two minutes by the new supercomputer.

BMW, Ford, Continental and Zenuity all are using DGX networks now for various applications. Nvidia also supplies a variety of in-vehicle processors and platforms for autonomous driving and other advanced technology development programs throughout the auto industry.