256 GPUs · air-cooled
32-node NVIDIA HGX H200 SuperCluster
Supermicro 32-node NVIDIA HGX H200 SuperCluster: 256 H200 GPUs, 36TB HBM3e, 400G InfiniBand spine-leaf, air-cooled, delivered as a validated rack-scale unit. Specs and UK/EU quotes.
256× H20032 nodes × 8 GPUs
36 TBHBM3e total
400G NDR1:1 GPU-to-network
9 racksair-cooled, validated
A cluster delivered as one product
Supermicro's SuperCluster packages 32 eight-GPU HGX H200 nodes with the spine-leaf NVIDIA Quantum-2 InfiniBand fabric, management nodes, storage connectivity and rack power, pre-cabled and pre-validated. You take delivery of a working 256-GPU cluster rather than 32 servers and a switching bill of materials. It scales in further 32-node units.
Who buys it
- Companies training models in the tens of billions of parameters and above, where the interconnect is the bottleneck.
- Cloud and inference providers standing up capacity in a European data centre.
- Research organisations replacing an ageing HPC cluster.
Before you ask for a quote
- Confirm the data centre can take roughly 40kW per rack air-cooled. If not, look at the liquid-cooled version, which fits in five racks.
- Decide on storage: most buyers pair it with a parallel file system, which we can quote alongside.
- Lead time on a full SuperCluster is longer than on single nodes. Ask early.
Get three quotes for the HGX H200 SuperClusterComparable pricing and lead times from authorised integrators, plus a cloud and hosted alternative so you can see the trade-off.
Request quotes