NVIDIA Grace Hopper Superchip Architecture In-Depth

The NVIDIA Grace Hopper Superchip Architecture is the first true heterogeneous accelerated platform for high-performance computing (HPC) and AI workloads. It accelerates applications with the strengths of both GPUs and CPUs while providing the simplest and most productive distributed heterogeneous programming model to date. Scientists and engineers can focus on solving the world’s most important problems. The NVIDIA Grace Hopper Superchip architecture brings together the groundbreaking performance of the NVIDIA Hopper GPU with the versatility of the NVIDIA Grace CPU, connected with a high bandwidth and memory coherent NVIDIA NVLink Chip-2-Chip (C2C) interconnect in a single superchip, and support for the new NVIDIA NVLink Switch System.