Following final yr’s Drive CX, Nvidia simply introduced an up to date model of its computing platform for self-driving automobiles, the Drive PX 2. In comparison with final yr’s mannequin, this can be a rather more highly effective beast, capable of course of knowledge from 12 video cameras and different sensors in actual time to make educated driving selections. The corporate calls it a supercomputer, and it’s the dimensions of a lunchbox.
Behind the scene, the Drive PX 2 options 12 totally different cores representing eight teraflops of calculation energy, 24 deep studying tera operations per second. The corporate is utilizing a 16nm structure, and it’s a hungry beast because it requires 250W of energy. Lastly, Nvidia is utilizing a liquid-cooling system. As a result of the PX 2 will probably be utilized in automobiles, that’s not an excessive amount of of an issue and Nvidia argues that automotive producers will simply plug the system into their present cooling options.
Nvidia co-founder and CEO Jen-Hsun Huang stated that the PX 2 was as highly effective as one hundred fifty MacBook Execs. The corporate is definitely evaluating GPU energy with the thirteen-inch MacBook Professional, which at present sports activities an Intel Iris Graphics 6100 chip. The PX 2 options two subsequent-gen Tegra processors, in addition to a Pascal-based mostly GPU. In complete, the system can push as much as eight teraflops and acknowledge as much as 2,800 pictures per second utilizing the AlexNet neural community-based mostly deep studying algorithm.
The corporate additionally introduced Nvidia Drivenet, its personal deep neural community. It has the equal of 37 million neurons and was educated one hundred twenty million objects thus far. Like different deep neural networks, it will get higher extra time.
Corporations will be capable of leverage this community, however Nvidia additionally insisted in saying that automotive makers would need to management their very own neural community.
The corporate is selling a platform strategy, which means that it needs to work with as many automotive makers as attainable to kickstart their self-driving efforts. Volvo goes to be the primary companion delivery Drive PX 2 in upcoming check automobiles. Nvidia has additionally partnered with Audi, Daimler, BMW and Ford to develop and check the PX 2.
As Huang confirmed throughout in the present day’s keynote, all of this energy is required to allow self-driving automobiles to know sufficient about their surroundings — and interpret it — to securely drive in visitors. In a single demo, Nvidia confirmed a brand new dashboard that may use this knowledge to point out drivers whether or not there are different automobiles round them at any given time. Because of this, “we quickly gained’t want rear-view mirrors anymore,” Huang joked. As Nvidia clearly famous, although, figuring out objects and planning primary paths is just the start. Wanting forward, making self-driving automobiles work in the actual world additionally signifies that automobiles can acknowledge circumstances as properly. Not each truck is identical, in any case — some are ambulances, for instance, and also you higher make area for these.