Today at its GTC event in San Jose, CA, Nvidia announced the latest mobility companies that have adopted its Drive Thor centralized car computer to power their next-generation consumer and commercial fleets. Intended for “new energy” vehicles and trucks to robotaxis, robobuses, and last-mile autonomous delivery vehicles, the next-generation AV SOC (system-on-chip) will integrate the new Nvidia Blackwell GPU architecture designed for transformer, LLM (large language model), and generative AI workloads, announced during Nvidia Founder and CEO Jensen Huang‘s keynote at GTC.

The Thor in-vehicle computing platform is architected for generative AI applications that are beginning to take hold in the vehicle industry as they have been in other consumer industries. The successor to Drive Orin is said to deliver feature-rich cockpit capabilities and safe and secure highly automated driving all on a centralized platform.

“Accelerated compute has led to transformative breakthroughs, including generative AI, which is redefining autonomy and the global transportation industry at large,” said Xinzhou Wu, Vice President of Automotive at Nvidia. “Drive Orin continues to be the AI car computer of choice for today’s intelligent fleets, but now we’re seeing mobility leaders looking ahead to bring Nvidia Drive Thor into their next-generation, AI-enabled vehicle roadmaps.”

Nvidia says that Thor is poised to revolutionize the automotive landscape, ushering in an era where generative AI defines the driving experience.

“The transportation industry is going through a seismic shift, driven by breakthroughs in accelerated computing and now generative AI,” said Danny Shapiro, VP of Automotive at Nvidia, in a GTC preview for the media.

The Drive platform spans applications from hardware and AI software that runs in the cloud and vehicle, and it extends into the Omniverse platform for digital twins, connecting the 3D real and virtual worlds. With it, Nvidia is aiming to help automakers transform their workflows from design through manufacturing to driving.

“No other company in the world offers a unified platform designed to digitize every aspect of the vehicle lifecycle,” said Shapiro.

 

Thor SOC adds new Blackwell GPU

Drive Thor is the successor to Drive Orin, building on that platform’s adoption but delivering four times the performance, at up to a 1000 TOPS (trillions of operations per second), to be able to run a range of deep neural networks and other AI workloads required for safe autonomous driving. The integration of Nvidia’s new Blackwell GPU boosts its ability for transformer and generative AI workloads.

The design of Thor is tightly coupled with the Blackwell GPU architecture. Like Orin, Thor is a scalable platform with different chip variants, thermal characteristics, and performance levels, Shapiro explained. As a system-on-chip, Shapiro says it has CPU, GPU, and processor logic for functions like video encoding/decoding and deep learning. Its many processors are designed specifically for AVs (autonomous vehicles).

At GTC, several leading-edge EV makers are revealing their next-gen AI vehicle fleets powered by Thor.

One of the most prominent is BYD, which is expanding its collaboration with Nvidia to extend from the car into the cloud. In addition to selecting Drive Thor to power its next-generation EV fleets, BYD plans to use Nvidia’s AI infrastructure for cloud-based AI development and training along with Nvidia’s Isaac platform for factory robotics, creating digital twins and planning logistics and manufacturing workflows in Nvidia Omniverse. In addition to planning and operating virtual factories, it plans to extend the technology into retail using Omniverse to have digital twins and custom car configurators that transform the customer purchasing and support experience.

“This is a true end-to-end adoption of Nvidia accelerated computing, AI, and Omniverse for a broad range of automotive applications,” said Shapiro.

In addition to BYD, Nvidia announced several customers across the transportation sector adopting Drive Thor to power their next generation of consumer fleets.

On the passenger car side, the premium luxury car brand of GAC’s Aion called Hyper announced that Thor will power its next-generation EVs and will begin production next year. Xpeng has announced it will use Drive Thor for the AI brain of its fleets, the computer powering the EV maker’s proprietary XNGP. The AI-assisted driving system will enable autonomous driving and parking capabilities, driver and passenger monitoring, and a variety of other in-cabin features. Li Auto and Zeekr have already announced they’re gonna be building their future vehicle roadmap on Drive Thor.

Beyond consumer vehicles, Thor is engineered for the diverse needs of other transportation segments where high-performance compute and AI are essential for ensuring safe, secure driving operations including trucking, robotaxis, and goods delivery vehicles. A number of these mobility providers are leading the contingent at GTC.

Nuro, which is developing SAE Level 4 autonomous driving technology and vehicles for commercial use, is integrating its proprietary AI-first software and sensors with Nvidia’s Drive solution and networking hardware.

Plus, a provider of autonomous driving software solutions for trucking, has announced the Level 4 Superdrive leveraging the compute performance of Drive Thor to understand the world around the truck and make safe driving decisions.

Waabi is building a generative AI self-driving system leveraging Thor for their next-generation AV 2.0 based autonomous trucks. Company engineers are integrating Nvidia’s centralized compute platform within the Waabi driver to power safe and reliable self-driving trucks at scale.

WeRide, in cooperation with a new Tier One partner Lenovo Vehicle Computing, is creating level four autonomous driving solutions for commercial applications and using Drive Thor. Integrated within Lenovo’s autonomous driving domain controller is the AD1, which will be used for a range of urban-centered use cases where functional safety, redundant safety design, sensor fusion, and scalability are a must.

 

Enabling generative AI applications

While new Drive Thor SOC customers and the addition of the Blackwell GPU made headlines, what may be just as significant is how the combination and other technologies can advance vehicle capabilities in generative AI. Shapiro says that vehicles are evolving into intelligent companions, seamlessly blending technology and comfort to enhance the driving experience.

“ChatGPT’s influence—and getting businesses to rethink the way they develop and deliver products and services—was really the theme of the last year,” he said. “Now what we’re seeing is generative AI coming to market, and Nvidia being a driving force to enable this.”

As with the general market, generative AI is changing the transportation industry by enabling rapid content generation of all types of inputs and outputs, text, images, video, sound, and animation. He said that generative AI workloads can unlock all kinds of new possibilities, interfaces, and applications.

Several automotive partners leveraging generative AI will be spotlighted at GTC, their work spanning from the data center to large language model training and systems running and then deploying at the edge.

Cerence has announced an automotive-specific LLM that serves as the foundation for the company’s next-generation in-car computing platform that will run on Nvidia Drive. Unveiled at CES, it provides an integrated in-cabin experience leveraging Nvidia AI foundation NeMo for training and fine-tuning LLM models and applying the proper guardrails to ensure a great user experience. DGX Cloud is the development platform and Nvidia AI enterprise software optimizes that entire experience.

Wayve says it is helping usher in this new era of autonomy it calls AV 2.0. It is characterized by large unified AI models controlling multiple parts of the vehicle from perception and planning to control. The company’s GAIA-1 is a generative world model for AV development running on Nvidia and the Lingo open-loop driving commentator uses natural language to enhance the driving experience. In this example, the input is video from a car’s front-facing camera. It can annotate or explain what’s happening in the environment to ensure that the passengers and driver of the vehicle understand the actions of the autonomous system.

In 2023, Li Auto unveiled its multimodal cognitive model called Mind GPT built on Nvidia’s TensorRT. The large language model based on an open-source library serves as the basis for the EV maker’s AI assistant for scene understanding, generation of information, knowledge, retention, and reasoning capabilities.

Nio has also been working with Nvidia to launch its Nomi GPT for a number of in-car experiences inside the car including the Nomi encyclopedia Q&A, a cabin atmosphere monitor, and a vehicle assistant—the AIs ensure greater convenience, enjoyment in the car, and safety. The solution does basic speech recognition, recognition, and command execution, and uses deep learning to understand and process complex sentences and instructions inside the car.

Geely is working with Nvidia on intelligent cabin experiences and an edge-to-cloud deployment. It is applying generative AI and large language model technology to provide smarter, personalized, and safer driving experiences with natural language processing, dialogue systems, and predictive analytics for intelligent navigation. Lenovo is unveiling a new AI acceleration engine that will run on Drive, featuring an AI model engine and compiler tools that will enable the development and deployment of LLMs inside vehicles.

SoundHound AI is developing an in-vehicle voice interface using LLMs running on Nvidia, and the company is creating generative AI capabilities designed for both real-time and available when offline.

“This is key when you’re driving,” said Shapiro. “You want to be able to not require a connection to the cloud, but all these systems need to be able to run locally on the vehicle as well.”

The solution offers access to SoundHound vehicle intelligence, which delivers information, whether it’s settings in the car or other kinds of troubleshooting directly from the car manual and other data sources, with a natural speech, interactive solution.

Tata Consulting Services is building an automotive generative AI suite powered by Nvidia GPUs and a software framework to accelerate the design, development, and validation of software-defined vehicles leveraging all conversational large LLMs in the vehicle.

At GTC, MediaTek will be announcing four new SOCs leveraging Nvidia technology like the Dimenisty Auto cockpit series now coming to market. These SOCs will come with support for Nvidia Drive OS so customers can develop on MediaTek- and Nvidia-based SOCs to have a full range of products. They feature Nvidia RTX graphics, and the systems will be available for AI-powered infotainment, rear seat entertainment, and cockpit head-up displays.