NVIDIA announces DRIVE Thor with 2000 Teraflops of Performance
In addition to the RTX 40 series graphics card at last night’s GTC, NVIDIA also announced the super chip DRIVE Thor for automatic assisted driving systems. Driven by the higher 2000 teraflops of performance, NVIDIA’s new platform now not only has stronger autonomous driving capabilities but also can be responsible for the computing work of the cockpit entertainment system. NVIDIA said that this is due to the powerful performance provided by the combination of the new Grace architecture CPU and Ada Lovelace architecture GPU, which has been applied on the H100 platform of the NVIDIA Hopper architecture before.
Image: Nvidia
The advantage of DRIVE Thor is its 8-bit floating point (FP8) capability. Typically, developers lose neural-network accuracy when moving from 32-bit FP data to 8-bit integer format. DRIVE Thor features 2,000 teraflops of FP8 precision, allowing the transition to 8-bit without sacrificing accuracy.
NVIDIA mentioned that in the current vehicle system, different functions and devices require different SoCs for separate processing, but DRIVE Thor achieves centralized management. A single chip can manage all functions, which can improve efficiency and reduce power consumption.
Geely-owned automaker ZEEKR is confirmed to be the first customer of NVIDIA’s new platform. The first mass-produced vehicle equipped with DRIVE Thor is expected to be launched in early 2025.
