NVIDIA unveils DRIVE Thor; 2,000 Teraflop centralized automobile pc; Geely ZEEKR first to undertake


NVIDIA launched NVIDIA DRIVE Thor, its next-generation centralized pc for protected and safe autonomous automobiles. DRIVE Thor, which achieves as much as 2,000 teraflops of efficiency, unifies clever features—together with automated and assisted driving, parking, driver and occupant monitoring, digital instrument cluster, in-vehicle infotainment (IVI) and rear-seat leisure—right into a single structure for higher effectivity and decrease total system price.

DRIVE Thor replaces NVIDIA DRIVE Atlan (earlier put up) and would be the follow-on to DRIVE Orin, which is presently in manufacturing and delivers 254 TOPS of efficiency.


The following-generation superchip incorporates the superior AI capabilities first launched within the NVIDIA Hopper Multi-Occasion GPU structure, together with the NVIDIA Grace CPU and NVIDIA Ada Lovelace GPU. DRIVE Thor with MIG assist for graphics and compute uniquely allows IVI and superior driver-assistance methods to run area isolation, which permits concurrent time-critical processes to run with out interruption.

Out there for automakers’ 2025 fashions, it can speed up manufacturing roadmaps by bringing increased efficiency and superior options to market in the identical timeline.

Signaling the transportation business’s assist for this new supercomputing structure, Geely-owned automaker ZEEKR introduced it can combine DRIVE Thor on its centralized car pc for its next-generation clever electrical automobiles, beginning manufacturing in early 2025.

DRIVE Thor helps multi-domain computing, isolating features for automated driving and IVI. Sometimes, dozens of electrical management models are distributed all through a car to energy particular person features. With DRIVE Thor, producers can effectively consolidate many features on a single system-on-a-chip (SoC), which eases provide constraints and simplifies vehicle-design growth, leading to considerably decrease price, much less weight and fewer cables.

DRIVE Thor is the primary AV platform to include an inference transformer engine, a brand new part of the Tensor Cores inside NVIDIA GPUs. With this engine, DRIVE Thor can speed up inference efficiency of transformer deep neural networks by as much as 9x, which is paramount for supporting the large and complicated AI workloads related to self driving.

One other benefit of DRIVE Thor is its 8-bit floating level (FP8) functionality. Sometimes, builders lose neural-network accuracy when shifting from 32-bit FP information to 8-bit integer format. DRIVE Thor options 2,000 teraflops of FP8 precision, permitting the transition to 8-bit with out sacrificing accuracy.

The brand new superchip additionally makes use of the newest NVLink-C2C chip interconnect expertise, whereas operating a number of working methods. The benefit of the NVLink-C2C is its skill to share, schedule and distribute work throughout the hyperlink with minimal overhead. This equips automakers with the compute headroom and suppleness to construct software-defined automobiles which might be repeatedly upgradeable by way of safe, over-the-air software program updates.

DRIVE Thor is designed for the very best ranges of purposeful security. NVIDIA has invested greater than 15,000 engineering years into security throughout its full stack. NVIDIA says it’s the solely firm with a unified security strategy throughout its total system, from the info middle to the fleet.

The DRIVE Thor SoC and AGX board are developed to adjust to ISO 26262 requirements. The software program stack is designed for each ISO 26262 and ASPICE compliance. The Thor SoC and software program are additionally designed and produced in alignment with ISO 21434, which offers the pathway for compliance with regulatory safety equivalent to UNECE Regulation 155.

Current DRIVE Orin clients can benefit from the platform’s scalable structure to transition present growth efforts to DRIVE Thor. Builders can reap the advantages of their software program investments throughout a number of product generations as they design for future manufacturing roadmaps.


Leave a Comment