avatarby Nick HallAugust 23, 20160 comments

Processing overlord NVIDIA has revealed details of the processor inside the Drive PX2 supercomputer that will power the self-driving cars of the future. Its nickname is Parker and it’s a beast, but Tesla is still streets ahead.

NVIDIA revealed the Drive PX2 at CES earlier this year, but now it wants to show the pure computing power of the processor. The Drive PX2 has two Parker processors and two Pascal GEFORCE GPUs that can handle the increasing demands of the next generation autonomous cars.

Serious deep learning power

Parker can burn through 24 trillion deep learning operations a second, just think about that for a moment. In addition, it has 1.5 teraflops of power that can handle billions more deep-learning-based AI operations.

The new chip offers a 50-100% improvement over previous mobile processors when it comes to the CPU performance. The CPU architecture gets a pair of 64-bit Denver 2.0 CPU cores and four 64-bit ARM Cortex A57 CPUs. Those Denver 2.0 CPUs are seven-way superscalar processors that get better algorithms and energy efficiency than their predecessor.

The chip can also handle and encode 4K video streams at 60FPS, which is a tough test of any processor, and this is clearly a huge step forward for some quarters of the motoring world. The 256-core Pascal can support deep learning and run a number of high-resolution displays simultaneously. This is an increasingly important function.

Parker and the complex architecture of the Drive PX 2

Power has increased, but so has demand

With almost every electric car running a touchscreen in the centre console and an electronic instrument binnacle, the computing power to run them is simply an essential part of the user experience.

Eight virtual machines can run on Parker at the same time, too, so it can isolate the driver assistance, digital cluster, in-car entertainment and more. It can also communicate constantly with Cloud-based systems to run updates that make your car safer and smarter.

The car makers can choose to use Parker on its own, or employ NVIDIA’s Driver PX 2 system in its entirety, which should give them a solid platform for self-driving cars and all the electronics they’re likely to need.

Tesla is still a long way ahead

It’s impressive stuff. Tesla, though, can afford a condescending smile.

Parker simply isn’t in the same league as the DGX-1 platform that powers Tesla’s OpenAI system. NVIDIA makes that, too, and it has 170 teraflops of pure computing performance. So even though Parker is a big step forward, it still isn’t even in the same league as the processor that sits in the Model S and Model X.

Volvo has publicly committed to Parker for the next gen XCX90, while NVIDIA claims that 80 carmakers, tier 1 suppliers and research facilities have also signed up to use the supercomputer. So that’s a clear indication that Tesla will hold the trump card when it comes to computing power for a long time to come.

Comments
Show Comments