The NVIDIA GH200 Grace Hopper Superchip integrates a Grace CPU with a Hopper GPU using NVLink-C2C coherent interconnect, delivering unified memory access for HPC and AI workloads. With 624GB of total memory (96GB HBM3 + 528GB LPDDR5X) and breakthrough bandwidth, it eliminates data movement bottlenecks for scientific computing, AI training, and data analytics at exascale.
Complete technical specifications for the GH200.