Nvidia bolstered its ongoing commitment to autonomous vehicle (AV) technology powered by AI on Monday with the unveiling of a next-generation Nvidia Drive Atlan System-on-Chip with nearly four times the performance of its predecessor Orin processor.
In addition, Volvo Cars announced it will use Orin in next-gen models, with the first to be revealed in 2022—the Volve XC90. Orin is capable of 254 trillion operations per second (TOPS) while Atlan will reach 1000 TOPS when it first appears in 2025 model cars. Carmakers will be able to sample Atlan as early as 2023.
Also, Faraday Future announced it is deploying Orin in its flagship ultra-luxury FF 91 electric vehicle for 2022. The FF 71 and FF 81 will also be powered by Orin when they are expected to be available in 2023 and 2024.
Nvidia’s revenues from vehicle-related chips and software are a small percentage of total revenues when compared to its data center and gaming revenues, but will grow as Nvidia grows, said Danny Shapiro, senior director of auto at Nvidia.
Nvidia currently works with hundreds of partners through its Drive line of products, including carmakers and companies producing mapping or sensors or software, he said. They do so because “we’re the best,” Shapiro told reporters in advance of Nvidia’s start of its virtual GTC21 conference expected to reach more than 100,000 registered participants.
“We’ve developed an open platform with all the tools and libraries available that’s easy to use, flexible, robust and end-to-end,” he said. “This is not just about writing code for a car, but it’s data center to car and back. Our GPUs in the data center are the same architecture as the SoC (System on Chip) in the vehicle.”
Nvidia has invested billions in Drive software and hardware, he said, calling it a “long-term play …with huge benefits. We’re working on saving lives, making society better by freeing up time from congestion. There’s a lot of social good, but we do see an ROI and it’s an important business for us. There’s a long term payout.”
One reason the payout will be good is that Nvidia is working with carmakers including Mercedes Benz on a software-defined approach where carmakers will be able to charge customers for software upgrades that rely on chips like Atlan with plenty of processing headroom. In other words, as applications grow in complexity, Atlan will be able to handle the load.
Nvidia CEO Jensen Huang announced Atlan as part of a GTC21 keynote early Monday, along with a slate of other products heavily focused on Nvidia AI prowess. He called Atlan a “technical marvel, fusing all of Nvidia’s strengths in AI, auto, robotics, safety and Bluefield secure data centers to deliver safe, autonomous-driving fleets.” The event attracted 100,000 engineers and Huang wore his trademark leather jacket during recorded remarks.
Ramping up compute faster than competitors
Sam Abuelsamid, principal analyst at Guidehouse Insights, said Nvidia deserves credit for “ramping up its compute capability faster than anyone.” Over the past seven years of automated driving system development “It has become clear that these systems will require more compute power than was initially thought.”
He noted in an email to Fierce Electronics said power usage by Atlan “ will likely be significantly less than the several hundred watts to power a four-chip Orin like the Nio Adam computer. Power efficiency is crucial, especially in electric vehicles where excess power consumption can cut into range." Nvidia told Fierce Electronics that the Atlan architecture scales based on customer needs. Partners can design with active cooling at under 50 watts, and others can design with liquid cooling over 100 watts, Nvidia said.
Nvidia’s first AV processor was Drive Xavier with 30 TOPS, which is already running in production cars and trucks. Orin is expected to appear in vehicles in 2022, three years before Atlan.
Bluefield and more
Among Atlan’s components is Bluefield, a data processing unit (DPU) with advanced networking, storage and security services to support compute and AI workloads seen in AVs. It offers data center infrastructure-on-chip programmability with a security enclave to defend against breaches and cyberattacks.
In addition to Bluefield, Atlan will rely on new Arm CPU cores and a next-gen GPU architecture. There will be deep learning and computer-vision accelerators and 400 Gbps networking.
Developers of AVs will appreciate that Nvidia also announced the 8th generation of Hyperion that includes sensors, compute and software—all sold out of a box. Two Drive Orin SoCs are in the box for Level 4 self-driving, which process data from (also in the box) 12 exterior cameras, three interior cameras, nine radars, and one lidar sensor in real time for safe autonomous operation. It will be available later in 2021, Nvidia said. Pricing was not announced.
Robotaxis and NEVs
Nvidia announced six robotaxi companies that are using Nvidia Drive to develop Level 4/5 autonomous systems. They are Cruise, Zoox, DiDi, Oxbotica, Pony.ai and AutoX
The AV industry is currently at level 2-plus, while level 4 vehicles are due to arrive later in the decade. A fully autonomous vehicle can operate without human supervision in a geofenced space, such as a city. They are suited to robotaxi operations which rely on high-resolution sensors in addition to a supercomputing platform such as the Atlan SoC.
Zoox CTO Jesse Levinson said in a statement that work with Nvidia has allowed the company to get two orders more magnitude of computation done with the same amount of power in just over a decade.
Nvidia also said it is working with eight different new energy vehicle (NEV) makers that are relying on Nvidia Driver to provide centralized and software-defined compute with AI capabilities starting in 2022. They are NIO, R-Auto, Xpeng, Li Auto, IM, Canoo, Faraday Future and Vinfast.
Omniverse and RTX
Nvidia announced its Omniverse simulation tool will available this summer for Drive Sim 2.0 to be used by AV carmakers to generate datasets to train the perception of upcoming AVs. Simulations require physically accurate sensors, available with Nvidia RTX, as well as accurate recreations of real-world driving conditions.
BMW Group and Nvidia also announced they are using the Omniverse sim platform for planning complex manufacturing systems. Their virtual factory planning tool integrates planning data and apps and allows real-time collaboration globally.
“Omniverse allows us to scale to do all types of aspects of sims for auto, development of factories, EVs and dev testing,” Shapiro said. Creating a digital twin with a sim allows developers to see “what’s going on outside and inside the car.” He predicted it will be used to design individual vehicles as well as managing fleets of trucks.
“You’ll be able to teleport into a vehicle and see what sensors see,” Shapiro said. “Drive sim really takes a huge boost.”
Shapiro said that automakers are willing to pay to add higher powered chips and sensors to their vehicles. He used as an example Mercedes-Benz and its use of Orin and surround sensors on every model. “They are enabling the entire fleet to be software defined,” he said. “As their installed base grows, customers could pay a subscription or pay one-time charges, but revenue potential is increased, which more than offsets the initial investment.”