Nvidia’s GTC Will Unveil What’s Next for Automotive AI

Nvidia, the chip company that has been powering much of the artificial intelligence revolution in cars, will give the latest details on its autonomous driving platforms at its GPU Technology Conference (GTC) this week in San Jose.

Automotive sales generate less than 10% of Nvidia’s business, which is dominated by gaming and data-center computing. But within the fledgling automotive AI industry, the Santa Clara-based company is like an 18-wheeler in high gear. It claims more than 320 companies and organizations are working with the Nvidia Drive platform for vehicles, and companies including Uber, Volkswagen and Chinese Internet and AI giant Baidu are developing self-driving cars with Nvidia.

Between the proliferation of sensors feeding data into vehicles, and drivers’ growing demands for information, video feeds and driver assistance features, cars need more computing power than ever.

Autonomous vehicles demand even more, and all that computing has to be done with as little power as possible, especially in electric cars that rely on batteries to move. The potential hardware and software market for this is huge. A report last year by Research and Markets predicted global autonomous vehicle revenue would grow to nearly $127 billion per year by 2027 — though it’s fair to say the technology and its acceptance could go many different ways by then.

Nvidia’s latest digital brain for vehicles is the Xavier system-on-chip, which the company has begun delivering to customers this year after spending $2 billion on research and development over four years.

But the company supplies more than silicon.

The Nvidia Drive platform combines hardware with software and development tools that car and app companies use to develop their own products. Xavier will form the heart of a series of hardware-software platforms for different vehicles and applications.

The star of this year’s GTC may be the Nvidia Drive Pegasus AI computing platform, the most powerful system based on Xavier. Nvidia calls it the first onboard AI supercomputer for fully driverless, Level 5 autonomous cars. The Pegasus includes two Xavier SoCs, plus two discrete GPUs, in a computer the size of a license plate.

Nvidia made its name with graphics processors and has expanded the capabilities of those chips into AI. Its systems are based on neural networks that can improve their analysis of data over time through “deep learning.”

For vehicles, that means better recognizing nearby signs and objects using streaming video and data from sensors around the car, as well as predicting events on the road more accurately. The onboard systems can be trained in a data center first and then continue to perform deep learning on the road. Nvidia trains its in-car platforms using its DGX supercomputer, which can simulate 60,000 miles of driving in one hour, drawing on Nvidia’s experience in gaming, the company says.

But Nvidia’s AI technology isn’t all about self-driving cars. At CES, the company also announced software for in-car virtual assistants and augmented reality, which may hit the road much sooner.

Nvidia’s main automotive rival is Intel, which acquired computer vision, mapping and other technology when it bought the startup Mobileye last year for $15 billion. At CES in January, where Nvidia unveiled Xavier, Intel said it could build a more complete and efficient autonomous car platform than Nvidia’s using the Mobileye assets and its own Atom mobile processor.

Some experts say it’s too early to compare those systems, which won’t appear in cars at the same time anyway. What Nvidia reveals at GTC may shed more light on how that rivalry, and the automotive advances that it inspires, will play out in the coming years.

— Stephen Lawson is a freelance writer based in San Francisco. Follow him on Twitter @sdlawsonmedia.

Leave a comment

Your email address will not be published. Required fields are marked *