Nvidia: Technology Leadership In Advanced Driver Assistance (NASDAQ:NVDA)

Semiconductor Maker Nvidia Reports Quarterly Earnings

Justin Sullivan/Getty Images News

Nvidia (NASDAQ:NVDA) has invested many years and billions of dollars developing systems-on-chip that combine ARM CPU cores with Nvidia’s GPU technology. Until recently, Nvidia struggled to find markets for its SOCs, even though the ARM architecture powers virtually all mobile devices. Around 2017, Nvidia pivoted its SOC business to what was then a new application, autonomous driving. While true vehicle autonomy has proved elusive, Nvidia’s partners are now able to offer driver assistance systems that don’t require constant driver monitoring. One of these, the Mercedes-Benz Drive Pilot system, is the first ever to be approved by the German government.

Mercedes-Benz introduces Drive Pilot, the world’s first Level 3 driver assistance system approved by the German government.

Last December, Mercedes-Benz announced that its Drive Pilot system was the first to be approved by the German Federal Motor Transport Authority. Drive Pilot is considered a “Level 3” Advanced Driver Assistance System (ADAS), under the SAE definition. Drive Pilot doesn’t require the driver to constantly supervise the system and be ready to take control, unlike other ADAS systems.

SAE ADAS Levels

SAE

The Mercedes announcement summarized how the system works:

On suitable highway sections and where traffic density is high, DRIVE PILOT can offer to take over the driving, initially up to the legally permitted speed of 37 mph. The controls needed for this are located in the steering wheel rim, on the left and right above the thumb recesses. When the driver activates DRIVE PILOT, the system controls the speed and distance, and effortlessly guides the vehicle within its lane. The route profile, events occurring on the route and traffic signs are correspondingly taken into consideration. The system also reacts to unexpected traffic situations and handles them independently, i.e. by evasive maneuvers within the lane or by braking maneuvers.

Drive Pilot uses an extensive suite of radar, LIDAR, ultrasonic, and video camera sensors:

Drive Pilot Sensors

Design News

Green lights on the steering wheel indicate when the conditions are met to enable Drive Pilot in Level 3 mode:

Drive Pilot Steering Wheel

Mercedes-Benz USA

Drive Pilot can also be operated in Level 2 mode when all conditions are not met, such as highway speed. So features such as lane keeping and emergency braking are still available, but with the driver required to be ready to take control.

Dan Carney, the author of the Design News article referenced above, described what it was like to let Drive Pilot do the driving:

Circulating the oval track at the Mercedes-Benz Immendingen proving grounds in the company’s EQS electric full-size sedan, Mercedes engineers did their best to encourage me to play Tetris on the car’s infotainment display even though I was seated behind the steering wheel.

The Mercedes engineers took the opportunity to demonstrate various capabilities of Drive Pilot:

Later, while cruising along in slow-but-steady traffic, another Mercedes driver swooped past, aggressively cutting across the bow of the EQS before continuing to the adjoining lane, as a driver who is about to miss their exit might do. Less-capable driver assistance systems would lose their cool in this situation, braking with hysterical fear at the sudden appearance of a vehicle so close in front. But Drive Pilot is smart enough to recognize a passing intruder in the car’s personal space and gives the offender a second to be on the way.

It says a lot about the confidence Mercedes has in the system that they were willing to take a journalist on a demonstration ride, albeit under controlled conditions. In contrast, Tesla’s (TSLA) Autopilot is still considered a Level 2 system by the U.S. National Highway Traffic Safety Administration.

The announcement said little about the underlying technology, but at CES in January, Nvidia revealed that it provided the “full stack” of hardware and software for Drive Pilot. At about 34 minutes into the presentation, Ali Kani, VP and General Manager of Automotive Business, states:

We have some partners that just buy our chips and core operating system, while developing their own software applications. Other partners, like Mercedes Benz, rely on us across this entire stack. From our self-driving software running on Nvidia computers in each car, to training AI models in the cloud, to synthetic data generation, to vehicle validation and testing of new features through simulation, which finally get pushed over the air to each Mercedes Benz vehicle.

Nvidia had officially become a trusted partner to a prestigious automotive brand in the most technologically advanced segment of the automotive supply chain.

Drive Pilot appears to be the only Level 3 system currently in production. Audi had planned a Level 3 system back in 2020, then scrapped it on liability concerns, according to CNET.

According to Electronic Design, Polestar’s next EV, Polestar 3, will feature a Level 3 system based on the Volvo Ride Pilot. Volvo has long been a partner of Nvidia in ADAS development and Ride Pilot is based on Nvidia Drive hardware. Polestar 3 is due to be unveiled this year.

BMW will offer Level 3 in the 2023 7 Series, according to Electronic Design, based on Mobileye technology. However, I’m a little skeptical of the timing and availability, since Mobileye’s processor that would be best suited for Level 3 is the EyeQ Ultra unveiled at CES in January. As I discuss below, the EyeQ Ultra isn’t set to enter production until 2025.

The Electronic Design article also mentions other very capable systems such as GM’s Ultra Cruise, Tesla’s Full Self-Driving Beta, and even Lucid’s DreamDrive. But none of these are expected to be Level 3 systems. Rather, they are “hands free” systems that still require the driver to be alert and attentive to the car and driving conditions, that is to say, not “eyes off.”

Powering Drive Pilot is Nvidia’s Orin system-on-chip, which eclipses the performance of Tesla and Mobileye and is in production now.

Nvidia unveiled the Orin SOC at its China GPU Technology Summit in December 2019. Nvidia promised that Orin would offer 7 times the AI capability of the previous generation SOC, dubbed Xavier.

Xavier was almost certainly a commercial flop. Although Nvidia had high hopes that Xavier would form the basis of an “off the shelf” automotive ADAS system to be marketed by component companies such as Bosch and ZF, it never came to pass.

Almost everyone had underestimated the difficulty of vehicle autonomy, and Tesla had announced its HW 3.0 SOC in April 2019. Tesla’s SOC was far superior to Xavier and set the bar for what was needed in hardware for a competent ADAS system, if not for full self-driving.

Nvidia’s answer was Orin, which arguably bested the performance of the Tesla SOC. I say arguably because performance claims for AI are usually measured in TOPS (tera operations per second or 10^12 ops/sec). But without specifying the exact nature of the operation, the TOPS metric becomes problematic to use.

At its Autonomy Day, Tesla specified the FSD chip to have a capability of 72 TOPS:

Tesla FSD chip

YouTube

Nvidia specifies Orin AI performance at 254 TOPS for 8-bit integer operations and that’s probably the most comparable to Tesla’s claimed performance:

Orin Developer Kit

Nvidia

Orin is certainly the larger chip with 17 billion transistors fabricated on a TSMC sub-14 nm process compared to the Tesla chip at 6 billion transistors fabricated on a 14 nm process. Much of the size of Orin is to accommodate various sensor and camera inputs and provide image signal processing.

Tesla’s FSD chip was a breakthrough in 2019, and is still used in Tesla vehicles, but it’s fallen far behind Nvidia. At Tesla’s AI Day Q&A, Elon Musk estimated that HW 4.0 would be 4 times the power of HW 3.0, which implies that HW 4.0 will be capable of about 280 TOPS, roughly comparable to Orin. But it’s not clear when HW 4.0 will arrive.

Also expected in the future is the Mobileye (INTC) EyeQ Ultra, that the company announced at CES in January:

EyeQ Ultra

Mobileye

Based on the specs in the chart, Orin is already more powerful than EyeQ Ultra.

For now, Nvidia’s Orin reigns supreme as the most powerful single chip solution for ADAS that’s in production. And Nvidia isn’t standing still. Next on the development roadmap is Atlan:

Atlan SOC

Nvidia

Nvidia claims that Atlan will offer a factor of 4 improvement over Orin, or about 1000 TOPS for 8 bit integer matrix operations. Atlan will likely be the highest-performing SOC for ADAS applications and robotics when it goes into production in 2025.

Investor takeaways: The growth potential of Drive

The continual escalation in SOC performance for ADAS systems conveys that automakers and chip developers haven’t really found the level of chip performance that meets the needs of full autonomy. Nvidia and other companies will keep developing ever more powerful SOCs until they hit that level.

For now, Nvidia offers the most powerful SOC, and most importantly, a full stack of hardware and software to go with it in the Nvidia Drive platform, as shown above in the developer kit. Nvidia has been able to translate the superior performance of Orin into multiple design wins. In its March 2022 announcement of the production start of Orin, Nvidia highlighted its design win pipeline:

The company also announced that its automotive pipeline has increased to over $11 billion over the next six years, following a series of design wins with vehicle makers from around the globe.

More than 25 vehicle makers have adopted the Nvidia DRIVE Orin™ system-on-a-chip (SoC). Starting this year, they are introducing software-defined vehicles, built on the centralized AI compute platform.

DRIVE Hyperion with Nvidia Orin™ serves as the central nervous system and AI brain for new energy vehicles – delivering constantly improving, cutting-edge AI features while ensuring safe and secure driving capabilities.

“Future cars will be fully programmable, evolving from many embedded controllers to powerful centralized computers – with AI and AV functionalities delivered through software updates and enhanced over the life of the car,” said Jensen Huang, founder and CEO of Nvidia.

Orin has been chosen by 20 of the top 30 passenger electric vehicle makers in the world to be their AI compute platform. In addition to Mercedes Benz, Jaguar Land Rover and Volvo Cars have also chosen Nvidia Drive.

The recent departure of Volkswagen AG CEO Herbert Diess over software delays suggests that more mainstream automakers will turn to Nvidia as well. The approach at VW was to try to do everything in house, as I discussed in a recent article for my Marketplace service, Rethink Technology.

These efforts included developing VW’s own hardware platform and operating system, called VW.OS. Not surprisingly, this led to delays in high end EVs under the Porsche, Audi and Bentley brands.

Automakers have fundamentally become systems integrators for the components and subsystems of their supply chains. Nvidia is perfectly positioned to feed into that paradigm as an automotive supplier.

Drive provides the right level of hardware and software so that automakers can configure and complete the Drive system according to their needs. Drive allows automakers to complete the high level software for the vehicle without getting lost in the details of operating systems or AI models. And automakers need to control the software at a top level in order to integrate the software and hardware as part of the production process.

VW apparently tried to emulate Tesla, but that’s probably not the right answer for most automakers. For most, the Mercedes/Nvidia partnership is the better model for achieving the programmable car of the future. I remain long Nvidia and rate it a Hold.