Supermicro SYS-821GE-TNHR (8× H200, 1,128 GB)

Enterprise 8U rackmount AI server featuring eight NVIDIA H200 SXM5 GPUs on an HGX baseboard, equipped with 1,128 GB of HBM3e memory, dual 5th Gen Intel Xeon Scalable processors, up to 8 TB of DDR5, 19 hot-swap NVMe bays, and delivering up to 32 PFLOPS of FP8 AI performance.

Designed to train and serve frontier-scale models on a single node, with NVLink and NVSwitch connecting all eight GPUs at 900 GB/s, and scale to multi-node clusters over NVIDIA InfiniBand or 400 GbE, shipping with NVIDIA AI Enterprise support and the standard CUDA, PyTorch, and Triton stack.

Rack-mounted and redundantly powered, with six 3,000 W Titanium supplies, roughly 7 kW of continuous draw, and the ability to run a 405-billion-parameter model at FP8 across all eight GPUs with full context.

Price available on request.