At CES 2016 in January, NVIDIA introduced NVIDIA DRIVE PX 2—its new generation high-performance computing platform for in-vehicle artificial intelligence applied to the complexities inherent in autonomous driving. (Earlier post.) This week, at NVIDIA’s GPU Technology Conference, the company announced that the DRIVE PX 2 platform will be used in the cars that will compete in the Roborace Championship, the first global autonomous motor sports competition.
DRIVE PX 2—which delivers processing power equivalent to 150 MacBook Pros—uses two next-generation Tegra processors plus two next-generation discrete GPUs, based on the Pascal architecture (earlier post), to deliver up to 24 trillion deep learning operations per second. Deep learning operations are specialized instructions that accelerate the math used in deep learning network inference. That’s more than 10 times the computational horsepower than NVIDIA’s previous-generation DRIVE PX.
Part of the new Formula E ePrix electric racing series, Roborace combines the intrigue of robot competition with electric racing.
Every Roborace will pit 10 teams, each with two driverless cars equipped with NVIDIA DRIVE PX 2, against each other in one-hour races. The teams will have identical cars. Their sole competitive advantage: software. It’s a contest to build the most advanced artificial mind.
|Roborace racer. Click to enlarge.|
The amount of information pouring into each of these autonomous high-speed racecars—and the need to make quick decisions—is incredibly demanding. That’s why Kinetik, the London-based investment firm behind Roborace, approached NVIDIA.
In addition to the 24 trillion deep learning operations per second, DRIVE PX 2’s multi-precision GPU architecture is capable of up to 8 trillion general purpose floating point operations per second—more than four times more than the previous-generation product. This enables partners to address the full breadth of autonomous driving algorithms, including sensor fusion, localization and path planning. It also provides high-precision compute when needed for layers of deep learning networks.