Driverless cars are very cool, but the technology that enables their self-driving smarts requires a novel mix of raw computational and deep learning power, using both general-purpose CPUs and GPUs to crunch real-time data and match it to a massive catalogue of imagery and vehicle profiles. A brand new Nvidia processor, the company says, has the power to finally make it happen.
The specific demands of processing the real-world information captured by front- and rear-facing cameras, radars, LIDAR and other ultrasonic sensors used in driverless cars requires a custom-built solution, and Nvidia’s Drive PX2 is already being used by more than 80 different car manufacturers.
Today, Nvidia shared some more details on the general-purpose processing side of the equation in Drive PX2. The supercomputer-in-a-box already includes two Pascal discrete graphics cards, but a brand new system-on-chip called Parker will handle everything from object processing to digital cockpits and in-car entertainment interfaces.
Built incorporating a low-power 256-core Pascal GPU to output graphics to in-car displays, the Parker processor has four general-purpose ARM Cortex-A57 cores — the kind you’d find in a 2015 or 2016 smartphone — but also includes two of Nvidia’s new 64-bit Denver 2.0 cores linked up through a “proprietary coherent interconnect fabric”, whatever that is.
The important real-world ramification of Parker is that it can be used with one chip or two, to scale to the power required for simple assisted driving, driver-aided autopilot or complete autonomous operation. The same scaling can also power more complicated in-car graphics and multiple in-car systems like digital instrument clusters and infotainment.
The entire setup can handle in-car video decoding streams of up to 4K and 60 frames per second, allowing for theoretical future vehicles to process a huge amount of visual data at high fidelity, allowing for clear recognition of different vehicle types without the slight lag associated with current self-driving and assisted driving systems.
Although most geeks associate the company with graphics chips for desktop and notebook PC gaming, Nvidia has its fingers in many different computational pies. Tesla, for example, already uses Nvidia hardware for the in-car entertainment in its Model X SUV. [Nvidia]
Gizmodo Australia is gobbling up the news in a different timezone, so check them out if you need another Giz fix.