GIGABYTE H233-V10-AAW1 barebone server NVIDIA GH200 Grace Hopper Superchip 2U 2-Node 8-Bay

2U dual-node AI barebones

The GIGABYTE H233-V10-AAW1 is a 2U rear-access server barebones housing two independent nodes. Each node is built around an NVIDIA GH200 Grace Hopper Superchip that pairs a 72-core Arm CPU with an H100-class GPU for giant-scale AI and HPC workloads. The system targets research labs and hyperscale operators who need dense coherent-memory compute in a standard rack footprint.

900 GB/s NVLink-C2C interconnect

The superchip replaces discrete CPU-plus-GPU boards by fusing them with a 900 GB/s NVLink-C2C link, eliminating PCIe bottlenecks between host and accelerator. Coherent memory spans up to 480 GB of LPDDR5X ECC on the CPU side and up to 96 GB of HBM3 or 144 GB of HBM3e on the GPU side per module. The jump is worth it when model size exceeds single-GPU memory and internode bandwidth must stay above 100 GB/s via BlueField-3 DPUs.

Four 10 GbE ports and eight Gen5 NVMe bays

The chassis supplies four 10 GbE ports through dual Intel X550-AT2 controllers and eight front-access Gen5 NVMe drive bays shared across the two nodes. Buyers who need more than eight hot-swap slots or who require on-board InfiniBand adapters should look elsewhere. The platform suits teams deploying NVIDIA GH200 NVL32 clusters where 32 superchips scale over NVLink Switch System fabric.