preface
144 high-performance cores with 1 TB of memory per second; doubles server chip performance and energy efficiency
Santa Clara, Calif.—GTC Conference—March 22, 2022 Pacific Time—NVIDIA today announced the launch of its first Arm Neoverse-based data center dedicated CPU for AI infrastructure and high-performance computing, delivering the highest performance, twice the memory bandwidth and energy efficiency of today's leading server chips.
The NVIDIA Grace CPU Super Chip consists of two CPU chips that are interconnected by NVLink-C2C. NVLink-C2C is a new type of high-speed, low-latency, chip-to-chip interconnect technology.
The GraceCPU Superchild is the first CPU-GPU-integrated "GraceHopper Superchip" module released by NVIDIA last year, which will be applied to large HPC and AI applications along with GPUs based on the NVIDIA Hopper architecture. The two superchievous chips use the same underlying CPU architecture and NVLink-C2C interconnect.
Jen-Hsun Huang, founder and CEO of NVIDIA, said: "A new type of data center has emerged, which is an AI factory that can process and refine massive amounts of data to achieve intelligence. The GraceCPU Super Chip offers the highest performance, memory bandwidth, and NVIDIA software platform in one chip, making it the 'CPU of the global AI infrastructure'. ”
Introducing the NVIDIA CPU platform
The Grace CPU Super Chip is built to deliver the highest performance, capable of accommodating 144 Arm cores in a single socket and achieving industry-leading analog performance in SPECrate2017_int_base benchmarks(1). According to NVIDIA Labs' estimates using similar compilers, this result is more than 1.5 times higher than the dual CPUs currently available on the DGX A100. (2)
GraceCPU Superchics also offer industry-leading energy efficiency and memory bandwidth, and an innovative memory subsystem based on LPDDR5x memory with error-correcting codes for the best balance of speed and power consumption. The LPDDR5x memory subsystem provides twice the bandwidth of traditional DDR5 designs, reaching 1 TB per second, while also significantly reducing power consumption, with the overall CPU plus memory consumption of only 500 watts.
Based on the latest data center architecture, Armv9, the GraceCPU Super chip offers the highest single-threaded core performance and supports Arm's next-generation vector scaling, which will bring direct benefits to many applications.
The GraceCPU superchip can run all NVIDIA computing software stacks, including NVIDIA RTX, NVIDIAHPC, NVIDIAAI, and Omniverse. The GraceCPU Super Chip, combined with the NVIDIA ConnectX-7 network card, can be flexibly configured into servers – either as a standalone pure CPU system or as a GPU acceleration server with one, two, four or eight Hopper-based GPUs, allowing customers to optimize performance for their specific workloads by maintaining only one software stack.
Designed for AI, HPC, cloud computing, and hyperscale applications
With the highest performance, memory bandwidth, energy efficiency, and configurability, grace CPU superchips stand out for the most demanding high-performance computing, AI, data analytics, scientific computing, and hyperscale computing applications.
The GraceCPU Superchip's 144 cores and 1TB/s memory bandwidth will deliver unprecedented performance for CPU-based high-performance computing applications. HPC applications are compute-intensive and require the highest performance cores, the highest memory bandwidth, and the most appropriate memory per core capacity to accelerate output.
NVIDIA is working with leading customers in HPC, supercomputing, hyperscale, and cloud computing around The Grace CPU Superchip. The Grace CPU Super chip and grace Hopper super chip are expected to be available in the first half of 2023.
disclaimer
Article source: NVIDIA
Where the source of information marked as other media sources, are reproduced from other media, does not mean that this network endorses its views, does not mean that this network is responsible for its authenticity, reprint please contact the original source. If you have any questions or doubts about the content of this article, please contact the background editor immediately, and the platform will quickly respond to you and process it. Indicate the company's original content, reprint please contact us