Watch more coverage of GTC Fall 2022.
NVIDIA has introduced a new automotive SoC that aims to serve as the central computing hub for self-driving cars.
The Santa Clara, Calif.-based company said the Drive Thor will output up to 2,000 TOPS of power with the company’s new FP8 data format, a major generational leap over its current Orin SoC lineup.
The chip features NVIDIA’s new “Hopper” multi-instance GPU to handle the machine learning workloads at the heart of autonomous cars, and also brings its latest “Lovelace” GPU architecture to the table. In addition, Thor adds a powerful Arm-based “Grace” CPU. With 77 billion transistors, the supercomputer-class SoC is designed to unify the clusters of computer systems that drive modern cars on a single platform.
Danny Shapiro, head of NVIDIA’s automotive business, said Thor is a step ahead of automakers who need high-performance hardware to enable more automated safety and self-driving features in their cars.
“Autonomous cars are one of the most complex computing challenges of our time,” Shapiro said. “With safety being paramount, no one is willing to release these vehicles into the wild until there is more computing power.”
NVIDIA is wrestling with Qualcomm (with its Snapdragon Ride Suite) and Intel (with its Mobileye EyeQ SoCs) to convince automakers to use its self-driving chips. The company has released multiple generations of its Drive SoCs capable of handling the safety-critical workloads that underpin automated and assisted driving. It also offers silicon to control dashboard displays, digital instrument clusters, camera mirrors and infotainment systems.
While a variety of different chips are typically required to drive all of these systems, NVIDIA says Thor has enough processing power to allow automakers to effectively consolidate many of their functions into a single chip.
NVIDIA signaled that Thor is still a work in progress for now. It will go into series production in 2024 and be on the road in 2025 vehicle models.
A new architecture
As software-enabled features become more of a focus for automakers, so does the hardware under the hood.
Today, more than 100 Electronic Control Units (ECUs) can be spread across a modern vehicle. Each module usually only has enough computing power to handle a single task, such as a parking assistance system. But as car complexity spirals out of control, automakers are shifting to “domain-based” architectures that combine many of these disposable modules into “domain controllers” that can be updated over time.
High-performance chips that each control unit is equipped with are designed to safely run several different functions at the same time instead of separate microcontrollers (MCUs). The systems run in separate software containers.
Other companies are transitioning to “zone-based” architectures, where a central on-board computer is connected via “gateways” to the sensors and other systems that transmit data over Ethernet in the car.
According to NVIDIA, Thor is suitable for centralized architectures in which many cameras, radar and sensors, and even displays are connected directly to the platform without the need for intermediary chips to preprocess data.
“We can do the sensor fusion directly instead of relying on an intermediary,” Shapiro said. But he added that Thor will give them the flexibility if their customers prefer to attach processors directly to the sensors.
Performance = Security
Packing everything onto a single chip requires a massive amount of processing power, which Thor promises to deliver.
Thor uses an automotive-grade version of Arm’s high-performance Poseidon CPU cores, designed for data centers, giving it access to one of the most advanced central processing cores on the market. The chip delivers up to 8x the performance of NVIDIA’s Orin SoC to process the massive amounts of data from autonomous cars’ cameras, radar and other sensors and then plan a safe route on the road.
Last year, NVIDIA introduced a new automotive SoC called “Atlan” that was expected to offer 1,000 TOPS performance on INT8 when it launched in 2024. However, the company announced that it has dropped the Atlan chip in favor of Thor.
While NVIDIA doesn’t reveal many details about its architecture, Thor will likely share many of the same building blocks as Atlan. But it also leverages many of the latest capabilities of the company’s GPUs.
Thor includes a new inference engine designed specifically for “Transformers”. This is a new breed of machine learning technology that is rapidly replacing Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) for AI. Transformers process video as a single frame, giving Thor the ability to process more data over time.
The Transformer engine — a new component in the tensor cores at the heart of NVIDIA’s server GPUs — can improve the performance of Transformer-based machine learning models by up to 10x, the company says.
However, when you integrate all of these disparate functions into a single architecture, you need secure hardware isolation to prevent security-critical and non-security workloads from interfering with each other.
To safely consolidate these disparate systems, the Thor SoC is also capable of domain-based computing. Therefore, it can partition itself so that safety-critical workloads can run without interruptions or delays.
In addition, the technology allows the chip to run multiple operating systems simultaneously. For example, the car’s core operating system could run on Linux while the digital dashboard runs on QNX or even Android.
This also opens the door for customers to funnel all of Thor’s power into the autonomous driving pipeline, or use a portion (perhaps 1,000 TOPS) to power the dashboard display and use the rest for ADAS.
NVIDIA said Thor and the AGX system based on it are both designed to meet the ASIL-D standard for functional safety under the ISO-26262 standard. The software stack is both ISO 26262 and ASPICE compliant.
The hardware and software is also designed to meet automotive industry safety standards, including ISO 21434.
While Thor will likely cost more than Atlan, NVIDIA said its customers should be ahead of the game when it comes to system-level cost savings as the car’s electrical and electronic architecture is simplified with a single SoC.
“You can imagine huge savings in terms of cost, reduced cabling, reduced weight and overall reduced energy consumption,” Shapiro said in a briefing with reporters. “Then there is the possibility of bringing new functions to these different ECUs with a single software update.”
Additionally, the absence of cables reduces the need for connectors that can pose a safety hazard if they become detached.
One SoC, many configurations
NVIDIA plans to launch various configurations of the Thor SoC, ranging from a single-chip solution to a “superchip” that connects two Thor SoCs using its NVLink C2C interconnect technology to run a single unified operating system. The company said it would give automakers the leeway to continually upgrade their cars over time with new services and even additional safety features via over-the-air updates à la Tesla.
“We will have a number of different options available so customers can select the right level of performance for their needs,” said Shapiro. “But it’s up to them to decide which configuration is right for them based on their needs and the sensors on their cars.” According to the company, Thor is designed to work with everything from advanced driver assistance systems (ADAS) such as lane change assistants to can be scaled to fully autonomous driving.
Chip energy efficiency is a key requirement for electric cars, where they must compete for limited battery life. According to NVIDIA, Thor is three times more efficient than Orin without sharing the specific numbers on performance.
Thor is coupled with the same Drive Software Development Kit (SDK) as Orin, which in conjunction with its scalable architecture enables companies port their previous software development to the new platform.
Watch more coverage of GTC Fall 2022.