Supermicro HGX 8× H200

Enterprise 8U rackmount AI server with eight NVIDIA H200 SXM5 GPUs on an HGX baseboard. 1,128 GB of HBM3e, dual 5th Gen Xeon, up to 8 TB of DDR5, 19 NVMe bays. Up to 32 PFLOPS of FP8.

Designed to train and serve frontier-scale models on one node. NVLink and NVSwitch link all eight GPUs at 900 GB/s. Scales to clusters over InfiniBand or 400 GbE. Ships with NVIDIA AI Enterprise, CUDA, PyTorch, and Triton.

Rack-mounted with six 3,000 W Titanium supplies and roughly 7 kW continuous draw. Runs a 405-billion-parameter model at FP8 across all eight GPUs with full context.

Request a quote