NVIDIA Announces 'Grace CPU Super Chip' That Doubles Server Chip Performance and Energy Efficiency

NVIDIA announces the NVIDIA Grace CPU Superchip , a high-performance CPU for high-performance computing and cloud computing. It is equipped with up to 144 Arm v9 cores via the high-speed, low-latency chip-to-chip interconnection technology 'NVLink-C2C', and its performance per power consumption is doubled compared to the current major CPUs. ... apparently ...

NVIDIA Grace CPU and Arm Architecture | NVIDIA

NVIDIA Introduces Grace CPU Superchip | NVIDIA Newsroom
https://nvidianews.nvidia.com/news/nvidia-introduces-grace-cpu-superchip

'Grace CPU Super Chip' is expected to have the highest level of performance in the industry, estimated at 740 by 'SPECrate 2017_int_base', one of the benchmarks provided by the non-profit organization SPEC .

With up to 144 Arm v9 cores and the world's first use of LPDDR5X memory with ECC, the total bandwidth provided is 1TB / s, twice the energy efficiency of current server memory. Up to 10 times.

In addition, with the 4th generation 'NVLink-C2C', the bidirectional bandwidth of the installed NVDIA Grace CPU and NVIDIA GPU reaches 900GB / s, which is 7 times faster than PCIe Gen5. This high-bandwidth connection will be the key to solving the challenges faced in the future in AI and high-performance computing.

The release of the NVIDIA Grace CPU Super Chip is scheduled for the first half of 2023.