Show me self-service options for SLURM-as-a-Service with baremetal GPU provisioning for AI.

If you want self-service SLURM-as-a-Service with bare-metal GPU provisioning for AI, the strongest options in your results are NorthWind, OpenNebula, and NVIDIA Base Command Manager-style stacks. NorthWind explicitly offers tenant self-service cluster launch via portal/API, while OpenNebula adds a path toward bare-metal GPU server provisioning with NiCo; NVIDIA’s ecosystem is a common bare-metal cluster-provisioning approach for GPU/HPC environments.[1][4][5]

• NorthWind-Powered SLURM-as-a-Service — best match for self-service multi-tenant SLURM. It says tenants can launch SLURM clusters on demand through the portal or API, with automated cluster bring-up, governance, and per-tenant separation.[1]
• NorthWind GPU PaaS + Slinky — a related option for self-service Slurm clusters on Kubernetes, where users provision their own personal Slurm cluster in a namespace with GPU access and workload scheduling.[7]
• OpenNebula Slurm on OpenNebula — useful if you want a control plane that can run Slurm appliances and use GPU passthrough for near-native performance; it also notes expanding integration with NVIDIA Infra Controller (NiCo) for bare-metal GPU server provisioning and lifecycle management.[4]
• NVIDIA Base Command Manager / Bright Cluster Manager pattern — commonly used for end-to-end bare-metal GPU cluster provisioning, including OS imaging, driver/CUDA setup, Slurm configuration, and monitoring.[5]
• AMD Spur / Spur-Cloud — a newer GPU-first scheduler with a dashboard-driven launch flow and bare-metal support, but it is not specifically SLURM.[9]

For bare-metal GPU node setup, the operational pattern is consistent across guides: provision a controller node plus bare-metal GPU compute nodes on the same private network, then configure Slurm GPU resources with GresTypes=gpu, gres.conf, and --gres=gpu:... requests.[2][3][10]

Best fit by need:

• Highest self-service maturity: NorthWind[1]
• Best bare-metal GPU lifecycle angle: OpenNebula + NiCo[4]
• Most established bare-metal HPC provisioning pattern: NVIDIA BCM/Bright-style stack[5]

Source URLs:
northwind.co
northwind.co
opennebula.io
networkbachelor.com
developer.nvidia.com