This year at CES, NVIDIA has focused on self-driving cars for which it has designed the NVIDIA DRIVE PX 2 platform. It builds on the same idea, presented last year, that an in-car supercomputer should be the center point of vehicles’ ability to be aware of their surrounding and make the right decisions based on an accurate sensing of the environment.
DRIVE PX 2 comes with one big improvement: NVIDIA claims that it is 10X faster than its last generation car platform. This new platform is based on two Tegra processors, the same ones used in tablets for example, and two discrete GPU (graphics processors).
The performance gain may be mainly related to those two discrete GPUs that are built from a new NVIDIA architecture codenamed Pascal. To make a point, the company says that regarding brute computing power, it is the equivalent to 150 MacBook Pros… but it fits in a shoe-boxed sized volume.
DRIVE PX 2 would be capable of fusing the input of 12 video cameras and other sensors such as ultrasonic, radar and lidar.
NVIDIA provides not only the hardware but, more importantly, the software components in a package called DriveWorks, which is a key piece of research that few car manufacturers could put together.
And that’s why names such as Audi, BMW or Ford have been evaluating NVIDIA’s platform and seem to have a great interest in NVIDIA’s DIGITS deep learning system.