By Murray Slovick, Contributing Editor
To fully appreciate what Nvidia is trying to accomplish with its new level 5 autonomous vehicle platform, take a look in the back of some of the level 2 research and development vehicles running around various cities. What you will find is essentially a full data center’s worth of electronics stuffed into the trunk with racks of computers and server-class GPUs running deep learning, computer vision, and parallel computing algorithms. And that’s why self-driving car developers usually choose SUVs, minivans, or large sedans as their testbeds: they need as much space as they can get.
Now consider Nvidia’s new AI computer system, code named Pegasus. It is as small as a license plate but nevertheless extends the company’s DRIVE PX platform to handle Level 5 fully driverless robo-taxis without steering wheels, pedals, or mirrors.
Nvidia DRIVE PX Pegasus is powered by four high-performance AI processors including two of Nvidia’s upcoming Xavier SoCs, which feature an embedded GPU based on the Volta architecture, a custom eight-core ARM CPU, and a new computer-vision accelerator. In all, DRIVE PX Pegasus delivers over 320 Teraflops per second—more than 10x the performance of the NVIDIA DRIVE PX 2. Unlike the liquid-cooled 250W DRIVE PX2 introduced in 2016, Pegasus units will be air-cooled.
Pegasus is designed for ASIL D certification, the industry’s highest safety level. Automotive inputs/outputs for Drive PX Pegasus include CAN, Flexray, 16 dedicated high-speed sensor inputs for camera, radar, lidar, and ultrasonics, plus multiple 10Gbit Ethernet connectors. Its combined memory bandwidth exceeds 1 terabyte per second.
Pegasus will be available to Nvidia automotive partners in the second half of 2018.