OpenMetal Integrates NVIDIA Blackwell and H200 into v5 Server Lineup

The new server configurations leverage the architecture introduced in OpenMetal’s v5 platform earlier this month, utilizing dual Intel Xeon 6530P processors and DDR5-6400 memory. Rather than treating GPU acceleration as an isolated tier, these systems incorporate PCIe 5.0 connectivity and 40Gbps private networking to ensure uniform performance across CPU and I/O subsystems. CTO Jamie Tischart noted that the move provides customers with predictable monthly billing and full control over their infrastructure, avoiding the limitations often associated with metered, multi-tenant cloud clusters.

The RP6000 series supports up to two NVIDIA RTX PRO 6000 Blackwell GPUs, each equipped with 96GB of GDDR7 memory and over 24,000 CUDA cores, targeting inference and visualization tasks. For more demanding large-model training, the H200 line utilizes NVIDIA H200 NVL PCIe GPUs, offering 141GB of HBM3e memory per card. These systems also include a five-year NVIDIA AI Enterprise software subscription to support production-grade deployments.

Because the hardware is provided as single-tenant bare metal, users can deploy standard AI frameworks like PyTorch and TensorFlow without the overhead of a shared hypervisor. The machines are currently available in the company’s Ashburn, Virginia data center, with support for both standalone operation and multi-node cluster configurations.