Self-driving car technology, which has stirred concerns over driver safety, is rapidly becoming viewed as something that could actually add to road safety.
Nvidia, the Santa Clara based visual computing company, was careful to make that point Tuesday as it unveiled Nvidia Drive PX 2, a new supercomputer for cars it says will speed the way to self-driving vehicles.
Self-driving cars will revolutionize society, CEO Jen-Hsun Huang told a crowd of 400 automakers and others at the annual CES expo in Las Vegas.
Autonomous cars will bring increased safety, new convenient mobility services and even beautiful urban designs, Huang predicted.
The lunch-box sized computer, 10 times faster than a previous version of the device, will be available to car-makers in the fourth quarter of this year, the company said, and will first be deployed by Volvo, which will outfit and lease 100 SUVs with the supercomputer in the company s home town of Gothenburg, Sweden, next year.
It fits in the trunk of a car and can deliver up to 24 trillion deep learning operations per second. It has the processing power of 150 MacBook Pros.
The device can process the inputs of 12 video cameras, lidar (remote distance sensing using lasers), radar and ultrasonic sensors, Nvidia said.
A fast-learner, the Drive PX 2 can rapidly learn how to deal with road debris, erratic drivers and constructor zones, the company said. It can also deal with poor weather conditions, extreme darkness and difficult lighting conditions.
Beyond immediate recognition of road hazards, the graphics chip company hopes that thousands of hours of road time can be captured by a deep learning neural network residing on supercomputers in the cloud.
Huang said the company is leveraging deep learning and supercomputing to create the brain of future autonomous vehicles that will be continually alert.
Eventually, he said, they will achieve superhuman levels of situational awareness.
Photo: The Nvidia Drive PX 2 automotive supercomputer (Nvidia)
The post Nvidia launches supercomputer for cars appeared first on SiliconBeat.
