Graphics processor company Nvidia used this week’s CES 2016 mega-event to introduce its Nvidia Drive PX 2, billed by CEO Jen-Hsun Huang the”world’s first in-car AI supercomputer for self-driving cars.”
As you may guess from that introduction, the Drive PX 2, upgraded from last year’s first model, is designed to be the on-board supercomputer that helps self-driving cars navigate the roads — and get better over time, using buzzy deep-learning technology.
In other words, this would be the “brain” that lets self-driving cars, self-drive.
To do that, it needs to be really, ridiculously powerful. And Huang says that this new car supercomputer is up to the challenge.
“The computional capability of the Drive PX 2 is roughly the same as 150 Macbook Pros,” Huang says.
The Drive PX 2 has 12 CPU cores, capable of 8 teraflops of processing power and 24 deep learning TOPS, Huang says. Don’t sweat the details, it’s just industry jargon to say that it’s extremely powerful. Indeed, it needs liquid cooling just so it doesn’t overheat.
“All of this within the size of a school lunchbox,” Huang says.
Internet of Cars
It’s a vital advance, he says, because “self-driving is hard.” Whether you take the Tesla approach, of using self-driving technology as an assistive tool, or the Google route of making the car utterly and totally self-driving, you need a lot of computing power behind the scenes.
That’s because “self-driving cars are hard,” and need a lot of data to safely ferry people around, Huang says. You need exceptional processing power to “read” the road and make sure that these self-driving cars get better the more mileage they log.
The idea, proposed by Huang, is an “Internet of Cars,” as each Drive PX 2-equipped car would contribute its data back to a mothership, that would help every other Nvidia-powered car learn from the collective experience of the herd. Nvidia is calling that mesh of self-improving “brains” the “Drivenet.”
But this is right in Nvidia’s wheelhouse: As a leading provider of graphical hardware for gamers and researchers alike, Nvidia has a lot of expertise in building systems that can make sense of video input and make it something understandable by a machine.