NVIDIA is not just about gaming graphics cards. The company had other news up its sleeve, including in the autonomous vehicle space. During the GTC keynote, NVIDIA CEO Jensen Huang announced a system-on-chip (SoC) called Drive Thor. NVIDIA says it designed the chip using the latest advances in graphics and processing to provide 2,000 teraflops of performance, all while keeping costs low.
NVIDIA says Drive Thor can unify all of the vehicles’ various functions, including infotainment, digital dashboard, sensors, parking, and autonomous operation, for greater efficiency. Vehicles with the chipset will be able to run Linux, QNX and Android simultaneously. Given the sheer amount of processing power that autonomous vehicle operations require, automakers can even use two of the Drive Thor chipsets in tandem using NVLink-C2C chip interconnect technology to run a single operating system.
Additionally, NVIDIA claims that the SoC marks a significant advance in “deep neural network accuracy.” The chipset has a transformer engine, a new addition to the NVIDIA GPU Tensor Core. “Transformer networks process video data as a single perceptual framework, allowing the compute platform to process more data over time,” says NVIDIA. He noted that the SoC can increase the inference performance of the transformer’s deep neural networks by up to nine times, “which is critical to supporting the massive and complex AI workloads associated with autonomous driving.”
The SoC follows and replaces Drive Atlan. It will be used in vehicles going into production starting in 2025. The first customer NVIDIA has lined up is the EV brand, which is already using Orin chipsets for level 3 automation. In the meantime, NVIDIA has hired two other partners from Drive Orin: the automakers and QCraft.
All products recommended by Engadget are selected by our editorial team, independent of our parent company. Some of our stories include affiliate links. If you buy something through one of these links, we may earn an affiliate commission. All prices are correct at time of publication.