NVIDIA (NASDAQ:NVDA) introduced the NVIDIA DRIVE Thor, a centralized computer system designed for autonomous vehicles.
The next-generation superchip comes with AI capabilities originally introduced in the NVIDIA Hopper Multi-Instance GPU architecture, along with NVIDIA Grace CPU and NVIDIA Ada Lovelace GPU. It can achieve upwards to 2,000 teraflops of performance, and was custom build to unify intelligent functions, including automated and assisted driving, parking, as well as driver and occupant monitoring, digital instrument cluster, and in-vehicle infotainment and rear-seat entertain. All of this in a single architecture intended for improved efficiency and lower overall system cost.
“Advances in accelerated computing and AI are moving at lightspeed. DRIVE Thor is the superhero of centralized compute, with lightning-fast performance to deliver continuously upgradable, safe and secure software-defined supercomputers on wheels,” said Jensen Huang, founder and CEO of NVIDIA.
DRIVE Thor can support multiple GPU instances (MIG) for graphics uses and advanced driver-assistance programs. It will be available for 2025 automobile models, and give a bump to production roadmaps by bringing higher performance and advanced features to market in the same timeframe.
Typically, individual functions are shunted off to multiple different electronic control units. One may handle steering while another handles the transmission and a third, fourth and fifth may handle the air conditioning, light switches and door locks. DRIVE Thor’s multi-domain computing and isolating functions consolidate many functions on a single system-on-a-chip (SoC), which simplifies development of vehicle design and eases supply constraints. This means that it will cost less, weigh less and use fewer cables.
Read more: The Mugglehead technology roundup: forward progress edition.
Read more: GreenPower Motor Company overcoming barriers to electric truck adoption
Read more: The Mugglehead technology roundup: future transportation edition
The new design will be the first AV platform to use an inference transformer engine. A transformer model is a neural network capable of discerning context and meaning by tracking algorithmic relationships in sequential data. It’s a new component in Tensor Cores within NVIDIA GPUs. The DRIVE Thor is able to boost inference performance of transformer deep neural networks by up to nine times, which is necessary to provide support for the large and intricate artificial workloads that come with self-driving.
The transportation industry has taken note of the new supercomputing architecture. Geely-owned automaker ZEEKR indicated it would add DRIVE Thor onto its centralized vehicle computer for its intelligence electric vehicles, with production beginning in early 2025.
“ZEEKR users demand a luxury experience that includes the latest technology and safety features. NVIDIA DRIVE Thor will support our mission of providing cutting-edge technology that fulfills the needs of our customers and ensures ZEEKR remains at the forefront of tomorrow’s innovations,” said An Congui, chief executive officer of ZEEKR.
It’s designed for maximal functional safety, the company has thrown more than 15,000 engineering years into the safety across its full stack. Now NVIDIA is the only company with a unified safety approach across its entire system, ranging from the data center to the fleet. It’s also been built to replace the NVIDIA DRIVE Atlan and will be built to add onto the DRIVE Orin, presently being produced and delivering 254 teraflops of performance.