2,000 teraflops of performance, supporting AI workloads for autonomous vehicles.
Improving efficiency and reducing system costs in autonomous driving and infotainment.

NVIDIA has integrated functions such as autonomous and assisted driving and infotainment into a single architecture, simultaneously improving efficiency and reducing overall system cost.

On the 26th, NVIDIA unveiled NVIDIA DRIVE™ Thor, the next-generation centralized computer for successful autonomous vehicles.

Drive Thor is the first autonomous vehicle platform to integrate the inference transformer engine, a new component of the Tensor Cores in NVIDIA GPUs.

This engine can accelerate the inference performance of Transformer deep neural networks by up to nine times, which is expected to be of great help in supporting the massive and complex AI workloads related to autonomous driving.

Another advantage of Drive Thor is its 8-bit floating point (FP8) precision.

During the development phase, loss of neural network accuracy typically occurs when converting 32-bit floating point data to 8-bit integer format.

Drive Thor provides 2,000 teraflops of FP8 precision, enabling 8-bit conversion without loss of accuracy.

Teraflops is a calculation that measures the performance of a supercomputer.At speed, one teraflop means one trillion calculations per second.

This has resulted in improved efficiency and reduced overall system costs by integrating intelligent functions such as △automatic/assisted driving and parking △driver/passenger monitoring △digital instrument panel △infotainment (IVI) △rear-seat entertainment into a single architecture.


Teraflops is a calculation speed that measures the performance of a supercomputer. 1 teraflops means 1 trillion calculations per second.

The next-generation superchip also features cutting-edge AI capabilities first introduced in architectures such as the NVIDIA Hopper™ Multi-Instance GPU and NVIDIA Grace™ CPU.

Support for MIG (Multi-Instance GPU) for graphics and compute enables domain isolation for IVI and advanced driver assistance systems, enabling uninterrupted execution of time-critical concurrent processes.

Targeting car models slated for release in 2025, DriveThor is expected to accelerate production roadmaps by bringing improved performance and advanced features to market.

“DRIVE Thor provides a continuously upgradable, secure and successful software-defined supercomputer for vehicles,” said Jensen Huang, CEO of NVIDIA. “ZEEKR, the automaker under China’s Geely, plans to integrate DRIVE Thor into the central computer of its next-generation intelligent electric vehicle.”

one side,
DRIVE Thor replaces NVIDIA DRIVE Atlan™, which is currently in production and delivers 254 TOPS of performance.It is expected to be the successor to e;).