Last week, the Lenovo Vehicle Computing unit rolled off its AD1 automotive-grade Level 4 autonomous driving domain controller production line at LCFC Hefei, China, the biggest manufacturing facility of the Lenovo Group. The achievement means that the new supplier is one of the first automotive Tier Ones in the world to produce an Nvidia Drive Thor-based platform—and at an impressive 2100 TOPS (tera operations per second).

The Lenovo Vehicle Computing team has quickly developed autonomous driving technology that is aligned with its vision of “AI for All.” It says it is innovating its solutions and services to bring a more unified and coherent smart experience, driving a sweeping transformation in consumers’ lives. Since its inception last year by the PC and consumer device powerhouse, Lenovo says its Vehicle Computing business has gained traction among a broad range of partners, potential customers, and the automotive industry.

The AD1 is an automotive-grade domain controller purposely built for commercial scenarios of SAE level 4 autonomous driving, emphasizing high computing power, safety, reliability, cost-effectiveness, and minimal maintenance cost. Complemented by a range of sensor interfaces, the company says it paves the way for the start of production of driverless vehicles such as robotaxis, robotrucks, robobuses, and robosweepers.

In March, Lenovo announced that WeRide had chosen its domain controller for its next-generation autonomous driving technology platform. The robobus solution will be used for a range of urban-centered use cases for which functional safety, redundant safety design, fusion, and scalability are important.

“WeRide is the only technology enterprise to simultaneously hold self-driving licenses in China, the United States, the United Arab Emirates, and Singapore,” said Tony Han, Founder and CEO of WeRide. “This collaboration with Lenovo Vehicle Computing, along with Nvidia’s accelerated compute and AI expertise, will enable us to deliver enhanced autonomous driving products, solutions, and services to customers worldwide.”

The adoption of Level 4 self-driving vehicles is approaching a pivotal moment for widespread implementation, according to Lenovo. The end-to-end large model technology in autonomous driving has stimulated the demand for high-performance computing.

Lenovo’s intelligent driving domain controller product line uses the Nvidia Drive Thor intelligent-driving platform leveraging the new Blackwell architecture for generative AI, said Rishi Dhall, Vice President of Automotive Business at Nvidia. It incorporates FP4 and FP8 operations into its AI inference, ensuring greater efficiency, maintaining high precision (comparable to FP16), and resulting in a fivefold increase in the inference speed of transformer models.

The AD1’s AI (artificial intelligence) computing power comes in at an impressive 2100 TOPS at FP8/INT8, and its CPU computing power, based on a SPEC CPU2017 measurement, reaches 174. Its functional safety level is up to ASIL-D of ISO 26262 and its cybersecurity complies with the ISO 21434 standard. In January, Lenovo Vehicle Computing announced that it had received the ASIL-D rating from TÜV Rheinland as a milestone for the controller’s development.

ISO 26262 is a core international standard aimed at minimizing risks in electrical and electronic system failures in automobiles is becoming more important as cars increasingly become autonomous, connected, and electrified. ASIL-D is the highest level of the functional safety standard and has the most stringent development process requirements.

“Receiving the ISO 26262 functional safety management system certification is a significant milestone for us, demonstrating that we’ve established a comprehensive development process system that meets the highest automotive functional safety requirements and industry standards,” said Donny Tang, Lenovo Vice President and Head of Lenovo Vehicle Computing and Lenovo Research’s Vehicle Computing Lab.

In the future, Lenovo will expand its offering to cost-effective domain controllers tailored for the Level 3 and Level 2++ ADAS (advanced driving assistance system) market, with various configurations to provide solutions for safer and more convenient ADAS features. The company’s UltraBoost AI middleware is compatible with the domain controllers, further enhancing computational efficiency, operator enhancement, and task scheduling.