A look under the hood of Nvidia’s Parker SoC

August 30, 2016 // By Christoph Hammerschmidt
At the recent Hot Chips conference in California, Nvidia introduced its new Parker SoC for automotive applications. The System is at the heart of the DRIVE PX-2 platform that is said to suit perfectly the demand of automated driving and other cockpit applications. Nvidia lifted the bonnet of its new SoC and provided a glimpse into a highly sophisticated graphics and computing engine.

According to Nvidia, Parker supports features that currently belong to the most talked about topics in the automotive electronics realm: Deep learning, hardware-level virtualization, a hardware-based safety engine for reliable fault detection and error processing and a range of I/O ports for integration into the car’s electronic ecosystem. The Parker is built around Nvidia’s highest performing Pasal GPU architecture and the latest generation of Nvidia’s Denver CPU architecture, which also is said to be extraordinarily power-efficient – the latter being crucial in cockpit applications where OEMs do not exactly love to admit any cooling system that generates acoustic noise such as blowers. The combination of Pascal and Denver delivers a computing performance up to 1.5 teraflops for self-driving AI systems utilizing deep learning schemes.


256 CUDA engines deliver enough computing power
for sophisticated deep learning algorithms

Thanks to its CPU architecture that consists of two 64-bit Denver 2.0 cores paired with four 64-bit ARM Cortex A57 CPUs, the Parker system delivers 50 to 100% higher multicore CPU performance than other mobile processors. The Denver 2.0 CPU is a seven-way superscalar processor supporting the ARMv8 instruction set. It implements an improved dynamic code optimization algorithm plus additional low-power retention states for better energy efficiency. The interconnection between the two Denver cores and the Cortex A57 CPU complexes is achieved by means f a proprietary interconnect fabric.

Working in concert with equally Pascal-based supercomputers in the cloud, the software of Parker based self-driving cars can be updated continually. The same holds true information to improve the driving performance of the vehicle’s electronic driver. While already one Pascal GPU with its 256 cores has sufficient computing power to run deep learning inference algorithms for the self-driving capabilities, the system is also scalable. The Nvidia Drive PX 2 employs two Parker chips with two discrete Pascal GPU cores. This combo delivers an awesome performance of 24 trillion deep learning operations per second, running even