Skip to main content
Dedicated clusters are bare-metal GPU deployments reserved exclusively for your organization. Unlike on-demand GPU instances (shared, hourly), dedicated clusters give you single-tenant infrastructure with guaranteed capacity, on a contract you sign directly with the GPU owner Runcrate introduces you to.
NVIDIA H100 (80 GB), H200 (141 GB), B200 (180 GB), and B300 (288 GB). All with NVLink intra-node and InfiniBand inter-node networking. See Available GPUs for full specs.
Reserved capacity starts at 8 GPUs, or at a month for any size, and is priced by quote. For smaller or shorter needs, consider on-demand GPU Instances, with 1, 2, 4 or 8 GPUs per instance and no commitment.
The owner’s offer states the delivery date. Capacity that is ready can be live sooner; a cluster built to order takes weeks to months.
A month or longer, priced by quote for the term. Longer terms may offer better per-GPU pricing.
Usually. You can add nodes mid-contract if the owner has them, as your contract with them sets out. Tell us and we’ll ask the owner.
Anything. You have full root access to bare-metal servers. Common setups include Kubernetes, Slurm, Docker, and direct bare-metal access. Runcrate can assist with managed Kubernetes or Slurm if needed.
North America (US West, US East, Canada), Europe (Germany, Netherlands, UK), and Asia Pacific (South Korea, Japan, Singapore, India, Vietnam). Availability varies by GPU type. Contact us for current options.
At the price the GPU owner quotes, for your GPU type, cluster size and term. You pay the owner directly, as your contract with them sets out. There is no deposit, and Runcrate charges you no fee: the owner pays Runcrate a commission.
The owner’s uptime SLA and support are part of your contract with them, and each offer shows them. A complaint about delivery also goes to a person at Runcrate.
Absolutely. Many customers prototype on the Inference Engine or on-demand GPU instances, then move to a dedicated cluster when they need reserved capacity at scale. Our team can help plan the transition.
Describe the cluster at runcrate.ai/clusters/reserved, or email support@runcrate.ai with your GPU requirements. We take your exact spec to the GPU owners that fit, and verify your company only when you book.