Nvidia's new supercomputer is for self-driving cars
Nvidia started things off at CES 2016 with some impressive news, as usual. First up, they announced the Drive PX2, which is the successor to last year's Drive CX. Now, the PX2 according to Nvidia is as powerful as 150 Macbook Pros, that is in the size of an average lunchbox. Inside there are 12 CPU cores that have a combined eight teraflops and can do up to 24 deep learning tera-operations per second. Not to mention the fact that the device is also water-cooled.
The reason that we need Nvidia's supercomputer for self-driving cars is because the cars need a lot of processing power so it can handle all of the sensors, controls and learning so it can be fully autonomous on the road. Jen-Hsun Huang said that "self-driving cars are hard" and they need Nvidia's processing know-how to do the job. Additionally, the PX2 can "process the inputs of 12 video cameras, plus lidar, radar and ultrasonic sensors." The PX2 system will be deployed in some Volvo cars, wich is Nvidia's first partner.
More interestingly, Nvidia has already created their own self-driving reference platform called "Nvidia Drivenet" where the platform is already testing their own self-driving cars. The system has nine inception layers so it can train itself to "perceive things out in the world." According to Huang, it only took months for the system to recognise objects in real time. Audi is their current partner testing out the system, and Nvidia even claims that the system was able to read the German road signs "better than a human can. Additionally, BMW, Daimler and Ford are also using Nvidia's system to further develop their self-driving cars. Stay Tuned to TechDepenedent for more CES 2016 coverage. To see all of our coverage, search under the CES 2016 tags and categories.
Via: | The Verge |
Source: | Nvidia |
Comments