If you watched the presentation today, then you might have noticed the announcement of a CPU processor as well that would be 'Grace', yes that was next to the new Hopper GPU architecture.

The Grace CPU chip is optimized for high-performance computing and artificial intelligence tasks. The (what NVIDIA calls Superchip) package consists of two Grace processors, each of which has 72 cores.

Nvidia's first venture towards a specialized data center CPU is with the Grace CPU Superchip. The Arm Neoverse-based processor will have a massive 144 cores and a memory bandwidth of 1 terabyte per second. It consists of two Grace CPUs coupled through the company's NVLink connector – a configuration reminiscent of Apple's M1 Ultra.

The processors are packaged together as a unit. With a score of more than 740 points in the SPECrate2017 int base integer test, the Grace CPU Superchip outperforms the competition for the time being.These cores are based on Arm v9 in the structure set architectural iteration, and there are a total of two CPUs in the Superchip module, which results in 144 cores in total. These cores are surrounded by an amount of LPDDR5x with ECC memory that is currently unknown, but which operates at a total bandwidth of 1 TB/s.

“A new type of data center has emerged — AI factories that process and refine mountains of data to produce intelligence,” said Jensen Huang, founder and CEO of NVIDIA. “The Grace CPU Superchip offers the highest performance, memory bandwidth and NVIDIA software platforms in one chip and will shine as the CPU of the world’s AI infrastructure.”

The NVIDIA Grace CPU Superchip makes use of the NVLink-C2C cache connection, which has a bandwidth of 900 GB/s, which is seven times more than the PCIe 5.0 standard. The Grace CPU Superchip is expected to launch in the first half of 2023.