During a press conference at CES 2016, Nvidia CEO Jen-Hsun Huang announced a new computer, the Drive PX 2, designed to enable self-driving cars. The Drive PX 2 is a new version of the Drive PX computer Nvidia announced at last year's CES, but with much more power.
Drive PX 2 uses two Nvidia Tegra processors and another two graphic processors, encompassing 12 processing cores. This hardware lets it process 8 teraflops of data, or 2,800 images per second. Huang compared the processing power to 150 Apple MacBook Pros.
As part of the self-driving car equation, this computer needs to process imagery from multiple cameras, radar and LIDAR, laser-based sensor and transceivers, building a virtual world in real-time matching the computer's external environment. On top of that, Drive PX 2 then calculates reasonable paths for the car based on an algorithm, taking into account other traffic, lane lines, traffic signs and pedestrians, among the many other objects in the world.
Nvidia has put much of its focus into self-driving car enabling technology, also announcing at CES 2016 Drivenet, a neural network based on millions of images. With Drivenet, Drive PX 2 can learn to recognize objects in the world based on similar objects previously identified from visual imagery. Drivenet and Drive PX 2 are components of a deep learning network, which is becoming a foundation computing technology to not only enable self-driving cars, but power speech recognition and other ways in which computers interact with humans and the outside world.
Danny Shapiro, Nvidia director of automotive, said that as automakers began to test the previous Drive PX computer, they found that it wasn't sufficient to power a production self-driving car, so Nvidia increased the computing power substantially, and added liquid cooling, to create Drive PX 2.
Huang said that Volvo will be the first company to test the new Drive PX 2 computer, using it in a fleet of a hundred self-driving test vehicles.