NVIDIA has launched a lesser version of its Drive PX 2 in-car computer, designed for autonomous vehicle control, and has said that it will power Baidu’s autonomous vehicle mapping and control systems. The new Drive PX 2 system follows on from the first-gen unit, described as “a liquid-cooled supercomputer for the car.” The original unit combined two Parker CPU / GPU chipset units in order to deliver the sustained high computing performance needed for artificial intelligence, deep-learning systems necessitated for self-driving cars. The announcement that Baidu is using the second-gen Drive PX 2 unit follows earlier announcements that NVIDIA and the Chinese company were working together on a full self-driving system architecture utilizing cloud technologies together with artificial intelligence.
The new Drive PX 2 uses 10 watts of power and is half the size of the original, which was around the size of a briefcase. In order to shrink the unit, NVIDIA are using one Parker chipset module whereas the original used two. Each Parker unit contains two of NVIDIA’s powerful, 64-bit custom Denver cores, four ARM Cortex-A57 application cores and a Pascal GPU unit. The Denver 64-bit core is a second generation version of the same chipset that powered the HTC-built Nexus 9, which scores highly in single-core operations but does not benchmark as highly as many competitor chipsets on account of these being two cores rather than a more usual four. However, the Parker chipset uses NVIDIA’s “fully coherent heterogeneous multi-processor configuration,” in other words, the chipset can use as many cores as it needs to. Thanks to this combination of powerful application processors, NVIDIA claims that the Parker unit delivers between 50% to 100% better multi-core processor performance compared with competitors.
NVIDIA’s efforts should help the overall industry. The original Drive PX unit has already achieved commercial success, but it’s an expensive piece of equipment that needs the original vehicle to be adapted in order to cope with the power requirements and heat output. NVIDIA’s cut down version lowers these requirements (at the expense of reducing the performance), but should make it easier for manufacturers to develop and experiment with hybrid autonomous systems. NVIDIA is already well-known for its high performance gaming-optimized gaming chipsets, but has worked to adapt its GPU and CPU technologies for the self-driving and artificial intelligence industry.