HGX
NVIDIA's baseboard reference platform integrating 8 GPUs with NVLink, the building block for many AI servers.
Current numbers
80 → 288 GBPer-GPU HBM on shipping parts: H100 80 GB, HGX B200 180 GB, B300 288 GB, Vera Rubin 288 GB
24–35 min automatic; 43–57 min with manual stepsTime to replace a failed HGX node after an injected fault, end to end (SemiAnalysis ClusterMAX 3.0)
HGX 8; NVL72 72; Rubin Ultra NVL576 576/8 racks; Kyber NVL144 144/rack, NVL1152 1,152/8 racksscale-up (NVLink) domain size: HGX node, NVL72 rack, announced Rubin Ultra and Kyber