Nvidia Continues to Put the Pedal to the Metal in Self-Driving Chip Race
What do tech companies and pro athletes have in common? A lot of times, the effort and attention they devote towards becoming the best (or at least one of the best) in a field determines how far they get relative to peers as much as their natural talent and abilities.
Amazon.com Inc.’s (AMZN) Alexa voice assistant is a good case in point. Though far from the first voice assistant to launch and lacking the tremendous mobile reach claimed by Siri and Google Now/Assistant, the aggressiveness with which Amazon invested in Alexa R&D, launched and promoted an innovative Alexa-powered home speaker line and laid the groundwork for Alexa to have (for now) an unmatched developer ecosystem has made it a credible rival to Apple Inc. (AAPL) and Alphabet Inc./Google’s (GOOGL) assistant platforms.
To a large extent, Nvidia Corp.’s (NVDA) recent momentum in the autonomous driving hardware space is the product of a similar mindset. Though the first computing board for its Drive PX self-driving platform was only shown off in early 2015, Nvidia’s manic product launch pace, together with several technology strengths, have arguably led it to steal a march on rivals who have been providing self-driving and/or driver-assistance solutions for a much longer time.
Nvidia continued its all-out push on Oct. 10 by unveiling the Drive PX Pegasus, a next-gen computing board that the company claims is powerful enough to support Level 5 autonomous driving systems — that is, systems that can replace human drivers in all situations, and in theory can enable cars featuring no steering wheels or pedals.
Such cars are likely at least a few years away — software is the big challenge here — but in the meantime, Pegasus could find its way into many cars supporting Level 3 autonomy, which lets a car take over from a human on certain roads and in certain conditions. And perhaps in time, Pegasus could work its way into cars supporting Level 4 autonomy, which can take over from humans in most real-world driving situations. Nvidia, for its part, notes that more than 25 of its 225 Drive PX partners are “developing fully autonomous robotaxis.”
Nvidia claims that Pegasus, due to ship in the second half of 2018, can handle 320 trillion operations per second (TOPS). Context is important for such a claim: Nvidia is talking about deep learning operations that Pegasus is especially good at, rather than general-purpose computing power. Regardless, Pegasus’ performance easily outclasses the 24 TOPS delivered by the Drive PX 2, which Nvidia unveiled in early 2016 and that powers Tesla Inc.’s second-gen Autopilot systems. It also easily surpasses the 30 TOPS promised by the Drive PX Xavier, a smaller, less power-hungry board set to ship in Q1 2018.
Pegasus’ performance is made possible by the fact that it packs two of Nvidia’s next-gen Xavier system-on-chips (SoCs), each of which contain high-performance CPU cores and an embedded GPUs based on Nvidia’s new Volta architecture, as well as two “next-generation” standalone GPUs. The Drive PX 2, by contrast, features two Nvidia Tegra X2 SoCs and two standalone GPUs based on Nvidia’s Pascal architecture, which launched in 2016. The Drive PX Xavier contains just one Xavier SoC.
Nvidia shares have risen 2.3% to $189.67 following the news, and earlier in the day made a new record high of $192.95. The Pegasus news seems to have soothed fears about Nvidia’s competitive position that popped up two weeks earlier after CNBC reported that Tesla, with the help of former AMD Inc. (AMD) engineering exec Jim Keller and AMD intellectual property, is working on a custom processor to power a future autonomous driving system. That week also saw Intel Corp. (INTC) confirm that it’s supplying Xeon CPUs (typically found in servers), programmable chips (FPGAs) and other parts for the self-driving systems developed by Alphabet’s Waymo unit.
Overall, one has to be impressed at the pace at which Nvidia’s engineers have been cranking out new Drive PX boards, with each new model significantly more powerful or less power-hungry than its predecessor. Four autonomous driving boards have now been unveiled in less than three years. And along the way, Nvidia has also unveiled a few driver-assistance solutions. Specifically:
- Drive PX Parker AutoChauffeur, a Drive PX 2-based solution meant to handle “point-to-point travel.”
- Drive PX Parker AutoCruise, a smaller board said to handle functions such as “highway-automated driving” and HD mapping.
- AI Co-Pilot, a platform meant to assist a human driver by monitoring both the driver and a car’s environment. It can do things such as warn a driver when he or she seems distracted, and display alerts about road conditions and nearby vehicles.
In tandem with the Pegasus announcement, Nvidia launched Drive IX, a software development kit (SDK) for creating AI Co-Pilot-powered systems. It also showed off a simulator that uses eight of Nvidia’s DGX-1 servers (they contain powerful Nvidia server GPUs) to simulate 300,000 miles of driving within five hours.
In addition, Nvidia announced it’s working with logistics giant Deutsche Post DHL to “deploy a test fleet of autonomous delivery trucks” beginning in 2018. DHL joins a list of marquee Drive PX partners that includes Mercedes-Benz, Volvo, Audi, Volkswagen, Toyota and auto parts giants Bosch and ZF Corp. Most, if not all, of these companies will likely be much less enthusiastic than Tesla about replacing Nvidia’s hardware with a custom processor.
Intel, for its part, has BMW and auto parts giant Delphi as partners in addition to Waymo. In addition, its processors will power a Level 3 autonomous system going into the 2018 version of Audi’s A8 luxury sedan. And the chip giant recently spent $15 billion to acquire Mobileye, whose vision processors enable the driver-assistance (ADAS) systems deployed by many top automakers. Auto chip giant NXP Semiconductors NV (NXPI) , which is set to be acquired by Qualcomm Inc. (QCOM) and last year unveiled its Bluebox autonomous driving platform, is also in the mix.
On the whole, however, Nvidia’s design win momentum looks stronger than that of any rival. The ability of its GPUs to effectively handle demanding deep learning algorithms is likely one reason for this, and Nvidia’s track record in developing power-efficient SoCs may be another. But clients also have to be impressed by the pace of its product launches and more broadly, by the resources it’s devoting to self-driving platforms.
If there’s a weak spot for Nvidia in this fight, it’s that Waymo and Tesla, owing to their software lead, are ahead of the pack when it comes to making autonomous driving a reality. If Waymo is able to ink deals with big-name automakers — it’s currently just testing some cars with Fiat Chrysler — and sticks with Intel, that’s bound to hurt Nvidia’s market share.
But it’s not a given that Waymo will stay with Intel forever, especially if Nvidia can provide superior hardware. And in the meantime, some of Nvidia’s partners are making real progress towards supporting partial autonomy. As long as Nvidia keeps its foot on the pedal (pardon the pun), its Drive PX sales are well-positioned to surge over the next few years.