Nvidia Drive PX Pegasus
Nvidia
Nvidia is pushing its platform for autonomous driving to the next level: the computing company has showcased an AI computer for driverless robotaxis at the Nvidia developers’ GPU Technology Conference (GTC) in Munich. The new platform, codenamed Pegasus, expands the well-known Nvidia Drive PX computer platform. The AI system can control self-driving vehicles at Level 5. Nvidia PX Pegasus delivers over 320 trillion operations per second, outperforming its Nvidia Drive PX 2 predecessor tenfold.
Nvidia Drive PX Pegasus
Nvidia Drive PX Pegasus is powered by four high-performance AI processors, which means in practice two Xavier SoC processors and two GPUs based on the as-yet undisclosed Volta successor. The whole system is to be about the size of a US car number plate, i.e. approximately 30 by 15 centimetres, and consume less energy than its predecessors.
Last September Nvidia unveiled Xavier, a SoC based on Volta, at 2016 GTC Europe in Amsterdam. Xavier comprises eight custom CPU cores with ARMv8 architecture, 512 CUDA cores of Volta’s GPU architecture and dedicated processors for handling data supplied by vehicle sensors; it also has the capability to encode HDR videos in 8K. There is no information on Volta’s successor so far. Nvidia did not announce any further details at 2017 GTC Europe.
ASIL D Certification
Pegasus is certified in accordance with ISO 26262, the highest automotive safety integrity level (ASIL D). The platform provides interfaces for CAN (Controller Area Network), FlexRay, 16 high-speed sensor inputs for cameras, radar, lidar and ultrasonics, as well as multiple 10-gigabits-per-second Ethernet connectors. Its combined memory bandwidth is designed to exceed 1 terabyte per second. Drive PX Pegasus is designed for a power consumption of 500 watts.
Pegasus should be available for Nvidia’s automotive partners from the second half of 2018.