Specifications

Overview

The NVIDIA GB200 Grace Blackwell Superchip integrates two Blackwell B200 GPUs with an NVIDIA Grace CPU on a single module, delivering 384 GB of HBM3e across the pair at 8 TB/s per GPU. The Grace CPU provides a coherent, high-bandwidth memory host that eliminates the traditional PCIe bottleneck between CPU and GPU.

Each B200 GPU packs 208 billion transistors and delivers 20 PFLOPS of sparse FP4 AI compute, making the GB200 the foundational building block of the Grace Blackwell NVL72 rack and the B200 HGX platforms. It is the successor to the Grace Hopper (GH200) superchip.

QS Compute sources GB200 superchips and Grace Blackwell systems for AI training, inference, and HPC deployments across APAC.