Building a Multi-GPU Cluster with OpenMetal H200s

Q: Can I build a multi-GPU cluster with OpenMetal H200 servers?

Yes, OpenMetal builds dedicated multi-GPU clusters of H200 servers on a private 40 Gbps mesh, built to order for distributed training and large-scale inference.

Explore GPU servers

An all-H200 cluster targets the largest models and bandwidth-bound work, with each node carrying one or two H200 cards (141GB HBM3e each), dual Intel Xeon 6530P, 1TB DDR5-6400, and a 6.4TB NVMe data drive. Nodes connect over a private 40 Gbps mesh (4x 10 Gbps LACP-bonded) that carries gradients, parameter exchange, and dataset traffic from OpenMetal storage; east-west traffic is not metered.

Within a node, two H200s pool memory over NVLink (282GB). Across nodes, distributed jobs use data and pipeline parallelism over the private network rather than a shared GPU-memory fabric, so size per-node GPU memory for workloads that need tightly coupled GPUs. Frameworks include PyTorch FSDP, DeepSpeed, and Megatron.

Clusters can also be mixed with RP6000 nodes to route cost-efficient inference and training to the cheaper card. Every node is single-tenant bare metal on fixed monthly pricing with included egress.