NVIDIA announces the first CPU 'Grace' for data centers, 10 times faster than conventional systems based on Arm

At the technology conference '

GTC 2021 ' held online by NVIDIA, we announced the CPU 'Grace ' for Arm-based data centers. Compared to the existing system that combines NVIDIA's graphic chip and Intel's CPU, it is said to have achieved 10 times the speed.

NVIDIA Announces CPU for Giant AI and High Performance Computing Workloads | NVIDIA Newsroom

'Grace' is NVIDIA's first CPU for data centers. The name comes from Grace Hopper, a computer scientist who is also known as the developer of the programming language 'COBOL' and the discoverer of the events that led to the word 'bug'.

A highly specialized processor for high performance computing workloads, such as training for next-generation NLP models with over 1 trillion parameters, using existing x86 CPUs in close combination with NVIDIA GPUs. Produces 10 times faster performance than the system.

Already, the Swiss National Supercomputing Center and the Los Alamos National Laboratory of the US Department of Energy have announced plans to build the first supercomputer using the Grace.

in Hardware, Posted by logc_nt