CES 2016: Nvidia announced today the Nvidia Drive PX2. According to CEO Jen-Hsun Huang, it is essentially a supercomputer for your car, made of 12 CPU cores and four GPUs, all liquid-cooled. 
This equates to roughly 8 TeraFLOPS of processing power. It is as powerful as 6 Titan X graphics cards, and the company compares it to “about 150 MacBook Pro” for self-driving applications.
Naturally, that much processing power is needed only for an extensive open learning network of Nvidia – called DriveNet. DriveNet is essentially a way for computers to learn to recognize objects better.
A demonstration video showed how the car could detect objects in real time, without relying on processing from the cloud.
Nvidia reports that the “marriage” of GPUs and the CUDA processing technology can greatly accelerate learning processes (20 to 40 times). In other words, the training of a process that once would have required a month will need only one day.
The first company to implement the autonomous driving platform based on Nvidia Drive PX2 will be Volvo, while the following are expected to follow: Audi, Mercedez-Benz and Daimler.
