While that’s not literally true — yet — the graphics-chip company is determined that its technology will be under the hood for the future generation of self-driving vehicles.
“Whether a car, truck or shuttle, they’re all trying to accomplish the same task: replace a human behind the wheel,” said Danny Shapiro, Nvidia senior director of automotive. “There’s no way to write code to account for everything the car could encounter; the world is too random. The only way to enable the car to handle the near-infinite number of things that can happen is artificial intelligence. For that you really need a supercomputer in the car.”
Nvidia’s graphics processing units, which dominate the fields of video games and visualizations for everything from medicine to architecture, are supercomputers that excel at solving complex problems, he said. The GPUs also have become a foundation for deep learning — using artificial intelligence to master all sorts of fields — for companies like Google, Microsoft, Facebook and Amazon that use them in massive data centers.
“A CPU (central processing unit, the brains for most personal computers, tablets, laptops and smartphones) has maybe two or four cores, which you can think of as lanes on a highway carrying data traffic,” Shapiro said. “Our GPU can have over 3,000 cores, so it’s like a 3,000-lane-wide highway. Imagine the amount of information you can push down that highway.”
All that computing power is packed into Nvidia’s Drive PX 2, a lunchbox-size device that looks like the inside of a powerful data center server packed with inputs for data from cameras and other sensors.
“This is our AI car computer, which delivers as much computing power as the world’s fastest supercomputer from the year 2000, which would have been larger than a basketball court,” said Sean Wix, senior technical marketing manager, as he showed it off. “It now fits on something the size of a license plate.”
Nvidia created several prototypes by retrofitting existing vehicles, and it has a permit from California to take them on test drives. It uses the cars for training its artificial intelligence system. Just as with a teenage driver, “we began driving lessons in a parking lot,” Wix said, showing a video of the car navigating an orange-cone obstacle course. But the system quickly learned the ropes of driving and soon progressed to one of the world’s curviest streets, he said, showing a video of Urs Muller, Nvidia chief architect for self-driving software, doing a “Look, no hands!” drive down San Francisco’s Lombard Street.
Nvidia’s chips can help automakers’ systems learn to recognize obstacles and road signs. Working with Audi, for example, “we took images of stop signs, taken from different angles, different times of day, different weather conditions, sometimes with a tree partially in the way, to teach the system,” Shapiro said.
Its end game is to sell the Drive PX 2 to other companies making cars, who will then adapt it to their own needs and do further training of its AI. Nvidia hasn’t disclosed the cost of the supercomputer, but press reports have said a developers’ kit is $15,000, while presumably bulk orders are considerably less.
Drive PX2 is already on the road handling the Autopilot feature for Tesla cars. Volvo is now using it in the CX90, its prototype autonomous car. Bosch, the world’s largest supplier of auto parts, this month said it will use Drive PX 2 as the basis for its own “AI car supercomputer.” Paccar, a Washington state company that manufactures Kenworth, Peterbilt and DAF trucks, this month said it is using Drive PX 2 in a proof-of-concept self-driving truck. Audi, Mercedes-Benz and Ford have said they will use Nvidia technology in forthcoming cars.
Rival chipmakers Santa Clara of Intel and Qualcomm of San Diego are also avidly pursuing the self-driving market. Intel this month plunked down $15.3 billion for Israel’s Mobileye, which makes visual sensors for cars, and has allocated $250 million for investments in autonomous car companies. Qualcomm last year offered $47 billion to acquire the Netherlands’ NXP Semiconductors, in large part to beef up its position in the automotive market, it said. (That deal is on hold pending regulatory issues.)
“Intel, Nvidia and Qualcomm are all trying to muscle their way into the lead in the (self-driving) industry,” said Paul Cuatrecasas, CEO of Aquaa Partners, which helps tech companies with mergers and acquisitions. “Whichever company gets it right could potentially be at the heart of another generation of technological growth.”
Cuatrecasas said the chipmakers see personal computer and smartphone growth slowing or reaching a plateau, heightening their need to conquer new markets.
“Nvidia has the advantage of speed with its expertise in GPU-accelerated computing,” he said. “But Intel has deep pockets, which it has recently demonstrated through its proposed acquisition of Mobileye. In rapidly changing markets with a lot of startup entrants, the firepower to buy up new emerging technology companies quickly may give them an edge.”
Shapiro naturally had a different view. “There’s definitely a difference between deep pockets and deep learning,” he said.
Nvidia is not exactly impoverished. For the fiscal year ended Jan. 29, it had $6.91 billion in revenue, up 38 percent from the prior year, while profits were $1.66 billion, up 171 percent. Intel’s fiscal 2016 revenues were $59.4 billion, up 7 percent, with profits of $10.3 billion, down 10 percent.
Nvidia’s robust growth helped make it the best-performing stock in the S&P 500 last year. Its shares skyrocketed 238 percent. The No. 2 performer, natural gas company Oneok, saw shares rise 135 percent.
©2017 the San Francisco Chronicle Distributed by Tribune Content Agency, LLC.
