Qualcomm Enters the Autonomous-Car Computing Ring with Snapdragon Ride

A few years ago, I attended Nvidia’s GPU Technical Conference, and during its Steve Jobs–ian keynote address, Nvidia founder Jenson Huang invited Elon Musk onto the stage for a friendly chat about technology. But by August of last year, the valentines floating back and forth that day had ended. After severing ties with Mobileye, who were the computing supplier of Autopilot’s HW1 (Hardware 1), Musk then jettisoned Nvidia, who’d been supplying the electronics for HW2 and HW2.5, too. The substantially more powerful HW3 (needed to absorb the data firehose that accompanies Full Self Driving) would now be designed in-house by Tesla, and in an especially Alpha moment of Silicon Valley chest-beating, claimed it could calculate at a rate of 144 TOPS (trillion operations per second)—clobbering the 21 TOPS of Nvidia’s Drive Xavier. Nvidia responded with a polite retort to those claims: Drive Xavier can do 30 TOPS, and moreover that’s the wrong comparison anyway. Drive Xavier was aimed at simpler, Level 2+ computing; HW3 should be compared to Nvidia’s 320-TOPS Drive AGX Pegasus—which it had specifically created for fully autonomous driving. (Moreover, Xavier itself can be goosed to 160 TOPS by adding a Pegasus GPU). Media critics, in turn, pointed out Pegasus’s high power-consumption (notably, the more recently introduced Drive AGX Orin can reline at 200 TOPS, with lower power demands).

I’m walking you through this arcane preamble of strange-sounding metrics to set our scene: although there’s a long list of companies chasing the AI smarts for autonomous driving, only Mobileye (Intel), Nvidia, and Tesla are making their own computing hardware. And the last two are engaged in a public performance war. Well, now the number of players has grown to four.

During CES 2020, Qualcomm hurled its hat into the ring—big time—with Snapdragon Ride. Qualcomm explains that it’s built on “scalable and modular heterogenous high-performance multi-core CPUs, energy-efficient AI and computer vision (CV) engines, and industry-leading GPU. The platform, with a combination of SoC (System on a chip) and accelerator can be used as needed to address every market segment offering industry-leading thermal efficiency, from 30 TOPS for L1/L2 applications to over 700 TOPS at 130W for L4/L5 driving.” It’s expandable—in daisy-chain fashion—and notably, is air- (not liquid-) cooled. Also eye-catching is that Qualcomm appears to be resisting the use of LiDAR—the first player to publicly side with Musk and his bet on radar and computer vision.

Qualcomm also sprinkled in a few other tech delicacies into its announcement—notably its dual, LTE-modem Snapdragon S820Am Automotive Platform that’ll appear in the Land Rover Defender 90 and 110 (allowing for simultaneous over-the-air software updates and streaming entertainment), and the C-V2X (cellular vehicle to everything) reference platform that extends the connected car into roadside infrastructure. But it’s Snapdragon Ride (available for testing in the first half of 2020 and scale production in 2023) that’s in the autonomous-tech spotlight as the San Diego challenger takes on the Silicon Valley Big Three. Make room, Nvidia, Mobileye, and Tesla.

The post Qualcomm Enters the Autonomous-Car Computing Ring with Snapdragon Ride appeared first on MotorTrend.

Comments are closed.