OpenMetal Introduces New GPU Servers for AI and HPC Workloads

OpenMetal Expands v5 Hardware Catalog

OpenMetal has officially announced the expansion of its v5 hardware catalog, introducing two cutting-edge private GPU server lines: the RP6000 and the H200. These new servers reinforce the company's commitment to supporting advanced AI training, inference, and high-performance computing (HPC) workloads.

Built on the robust NVIDIA RTX PRO 6000 Blackwell Server Edition and NVIDIA H200 NVL technologies, these servers offer substantial performance improvements and features designed to cater to modern computational needs. Each server is powered by dual Intel Xeon 6530P processors alongside a massive 1TB of DDR5-6400 memory, expandable to 2TB. This high-performance infrastructure is complemented by PCIe 5.0 connectivity, NVMe storage options, and 40Gbps private networking, ensuring that users benefit from exceptional responsiveness and reliability.

Jamie Tischart, OpenMetal's Chief Technology Officer, explained, "With the v5 hardware catalog, we have significantly modernized our bare metal and private cloud offerings. Adding the RP6000 and H200 reflects our understanding of the market's demands for dedicated GPU resources that can enhance AI and HPC tasks without unnecessary complexity or hidden costs. Our customers can expect transparent monthly billing and complete control over their infrastructure."

GPU Server Features

The RP6000 servers are equipped to support one or two NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. Each GPU comes packed with 96GB of GDDR7 memory, 1.79 TB/s of memory bandwidth, and 24,064 CUDA cores. These specifications allow for a wide range of applications, from inference serving to rendering and mixed AI tasks that benefit from high VRAM capacities.

Conversely, the H200 servers accommodate one or two NVIDIA H200 NVL PCIe GPUs, which are particularly designed for handling large-model training and memory-intensive inference workloads. They feature a remarkable 141GB of HBM3e memory alongside 4.8 TB/s of bandwidth. A notable advantage of the H200 is the inclusion of a five-year subscription to NVIDIA's AI Enterprise software, enabling firms to deploy AI solutions with a robust and supported framework within their dedicated environments.

Both GPU server lines operate as single-tenant bare metal infrastructures, which means there’s no shared hypervisor complicating operations. This facilitates straightforward deployment of common AI and machine learning frameworks, such as PyTorch, TensorFlow, JAX, and Hugging Face Transformers. Companies have the flexibility to implement these servers as standalone solutions or part of multi-node GPU clusters, all connected through OpenMetal's dedicated private network.

Availability and Future Plans

The RP6000 and H200 have already been made available at OpenMetal’s data center located in Ashburn, Virginia, with plans for expansion into additional locations soon. Additionally, teams interested in exploring the capabilities of these new servers have the option to access proof-of-concept deployments. This initiative allows organizations to validate their specific workloads and ascertain the server configurations that best meet their needs before making a purchase decision.

About OpenMetal

OpenMetal specializes in providing open-source cloud and infrastructure-as-a-service solutions. Their offerings include Bare Metal Dedicated Servers, private GPU servers and clusters, Ceph storage clusters, and hosted private clouds powered by OpenStack and Ceph. The company serves a diverse clientele, including hosting providers, SaaS businesses, managed service providers, and enterprises seeking dedicated infrastructure with transparent pricing and full operational control.

OpenMetal is paving the path for innovation in cloud infrastructure while ensuring that organizations have access to the computing resources they require to succeed in the rapidly-evolving digital landscape.