NVIDIA Unveils Grace CPU Superchip to Power Next-Gen AI Data Centers
NVIDIA introduces its first Arm-based data center CPU, the Grace CPU Superchip, which delivers industry-leading performance and energy efficiency for AI infrastructure. The new processor features 144 cores and doubles the memory bandwidth of current server chips.
NVIDIA announces the Grace CPU Superchip, its first Arm Neoverse-based discrete data center CPU designed specifically for AI infrastructure and high performance computing. This new processor delivers the highest performance available while providing twice the memory bandwidth and energy efficiency compared to current leading server chips.
The superchip packs 144 Arm cores into a single socket and connects two CPU chips coherently using the new NVLink-C2C interconnect. It utilizes an innovative LPDDR5x memory subsystem that achieves a massive bandwidth of one terabyte per second while keeping the total power consumption of the CPU and memory at just 500 watts.
Built on the latest Armv9 data center architecture, the Grace CPU Superchip runs all of NVIDIA's computing software stacks, including RTX, HPC, AI, and Omniverse. CEO Jensen Huang describes the processor as the ideal CPU for modern AI factories, positioning it as a complementary component to the previously announced Grace Hopper Superchip for giant-scale applications.