The AV market is more promising than potent, but that did not stop Jensen Huang from touting Nvidia's plan to corner the "multi-trillion-dollar transportation ecosystem."
Today, the autonomous vehicles (AV) market is more promising than potent. It still has a long way to go before growing into a volume business. But that uncertainty did not stop Jensen Huang, Nvidia’s CEO, from touting his company’s plan to corner what he describes as a “multi-trillion-dollar transportation ecosystem.”
Talking about AVs during a keynote speech at Nvidia’s Graphic Technology Conference (GTU) Monday, Huang said, “Twenty years ago, all of this was science fiction. 10 years ago, it was a dream. Today, we’re living it.”
Although all the speech seemed classic Huang bravado, a slew of Nvidia technology announcements Monday provided impressive substance.
Huang revealed DRIVE Atlan, scheduled for production in 2025 models. Nvidia’s next-generation DRIVE SoC will deliver 1,000 TOPS. The Atlan consists of a new CPU, GPU and deep learning accelerator (DLA) with the latest in networking and security necessary for next-generation AVs.
Also rolling out was DRIVE Hyperion 8, Nvidia’s AV platform designed for data collection, AV development and testing.
Huang also announced DRIVE Sim, Nvidia’s new simulation suite enabled by Nvidia Omniverse for digital twin — needed for running real-world scenarios for AV development and validation.
All in all, Nvidia claims to have an $8 billion automotive pipeline, including deals with companies such as Volvo Cars, Mercedes-Benz, NIO, SAIC, TuSimple, Zoox, Cruise, Faraday Future, VinFast and more.
Of all the technology and products unveiled by Nvidia Monday, Egil Juliussen, an independent auto industry analyst, believes the Hyperion 8 is possibly “the most important.”
The reason is that the AV development platform is likely to seed “new design-wins at the R&D phase from OEMs, startups and transportation companies.” He explained that the platform “greatly simplifies [the process of] getting started with an AV system, and developing future products.” This is akin to “a development system for the IC industry,” he said, which is always important. Add to this “cloud computing software for AI-based AV tech,” Juliussen noted, and it becomes “a powerful development system.” During the keynote, Huang mentioned separate agreements Nvidia has with AWS and with Google Cloud.
Drive Atlan: What does 1,000 TOPS mean?
Drive Atlan is “showing the continued performance advances in GPU-based SoCs,” said Juliussen. “It is a given that Nvidia has to do this on a regular basis.”
However, Mike Demler, senior analyst at The Linley Group, noticed that “Nvidia seems to be preannouncing future processors earlier and earlier with each generation.”
He said, “Orin isn’t even in production yet. but they’re now announcing Atlan as its successor?” At this point, “[Atlan’s] block diagram and 1,000 TOPS spec are just high-level design objectives,” Demler said.
When EE Times asked which process technology node Atlan will be using, Danny Shapiro, Nvidia’s senior director of Automotive, declined comment. Nor would Nvidia disclose when the SoC will tape out. Demler doesn’t believe the industry will see the first [Atlan] silicon until late 2023.
Demler sees that Atlan raises more questions than answers. “Nvidia says TOPS is the new horsepower, but what about efficiency? They keep moving up the AV performance requirements.”
He pointed out, “Remember when Drive Pegasus with Xavier @320 TOPS was an L5 system, then it moved to 400 TOPs with Orin, now 1,000 TOPS for Atlan? Mobileye is building AVs with about 1/10th the ‘TOPS’ rating, so clearly that’s not a reliable measure.” He added, “Mercedes and Volvo design wins are promising, but meaningless until we see them in production.”
Demler isn’t alone. Most analysts take a claim of 1,000 TOPS with a grain of salt.
Juliussen, for example, said, “Yes, 1,000 TOPS is good, but this is peak performance rarely achieved.” Juliussen previously noted, “I think TOPS stands for Totally Optimistic Processor Speed.” He added, “The other important factor is power consumption. It is more meaningful.” Nvidia is not yet disclosing that number.
Demler also questions Atlan’s SoC architecture. He said, “With Atlan, Nvidia aims to address instrument cluster, infotainment, ADAS/AV, driver-monitoring, network gateway… all in one chip.” Demler stressed that Nvidia is treating the car like a server on wheels, but unlike in a datacenter, the AVs don’t have unlimited power. “It’s not clear that putting all those function on one chip is the best approach, though they did say we could expect a family of Atlan chips.” He suggested that Atlan’s “all-in-one chip” description is designed more for headlines than actual designs.
That said, Nvidia’s strength with all the DRIVE SoCs is that they are “upward compatible,” Juliussen pointed out. That makes it possible for a software ecosystem to be re-used with each new generation.
Nvidia vs. Mobileye
How has Nvidia taken the lead in the automotive market? “It is a combination of the Nvidia eco-system (hardware, software, AI models etc.) plus the promise of future SoCs with higher performance,” according to Juliussen.
Juliussen believes Mobileye has “a system strategy.” It was explained in the Mobileye-Udelv announcement today. Mobileye’s assets are “REM, RSS, True Redundancy for sensors, which is very good and will get design wins at the AV use-case level.” In Juliussen’s mind, however, the question is “if Intel-Mobileye can evolve this system strategy to compete at a hardware component level when the production volume moves to 100,000’s-1 million’s levels.”
Nvidia’s multi-level design wins
Notably, Nvidia gets wins at multiple levels, Juliussen said.
“The first level is at the AV software platform level, where most of the leading players started out using Nvidia-based hardware and are likely to continue to use Nvidia’s SoCs.”
In his latest Egil’s Eye column, he pointed out: “Many of the AV software platform companies use Nvidia AV hardware including Aurora, AutoX, Nuro, Navya, Pony.ai, TuSimple, Yandex, Zoox and others.”
Huang in his presentation listed such robotaxis as Cruise, Zoox, DiDi, Oxbotica, Pony.ai and AutoX.
The next level for Nvidia’s design wins is the Tier 1s, said Juliussen. Bosch, Continental, ZF and others offer Nvidia-base automotive grade processor boards for the OEMs.
Waves of AI
During his keynote, Huang cited the “waves of AI” that have emerged over the decades.
Huang called Internet giants (i.e. Google, Facebook, Amazon, etc.) “the first adopters [of AI],” who have “excellent computer scientists, large computing infrastructures, and the ability to collect a lot of training data.”
The Nvidia CEO said, “We are now at the beginning of the next wave.” He said it will feature “enterprise and the industrial edge, where AI can revolutionize the world’s largest industries. From manufacturing, logistics, agriculture, healthcare, financial services, and transportation.”
Huang added to this next wave “autonomous systems,” among which “self-driving cars are an excellent example.”
Of all the AI waves crashing on technology’s shores, he sees the industrial edge and autonomous systems as “the most challenging, but also the largest opportunities for AI to make an impact.”