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

Here are the most relevant self-service options for SLURM-as-a-Service with bare-metal GPU provisioning for AI:

• NorthWind-Powered SLURM-as-a-Service: Offers fully managed, multi-tenant SLURM with self-service cluster launch via portal or API and automated cluster bring-up; it’s explicitly positioned as a cloud-like on-demand service.[1]
• NorthWind GPU PaaS + Project Slinky: Enables teams to provision their own Slurm cluster within a namespace on shared Kubernetes, with GPU access and workload scheduling through a console or APIs.[13]
• OpenNebula Slurm appliances + NVIDIA Infra Controller (NiCo): OpenNebula provides a Slurm Controller and Slurm Worker appliance flow, and says it is expanding toward bare-metal GPU server provisioning and lifecycle management via NiCo.[5]
• NVIDIA Slurm on Kubernetes: NVIDIA describes slurm-operator for running full Slurm clusters on Kubernetes infrastructure and slurm-bridge for integrating Slurm scheduling with Kubernetes workloads; paired with the NVIDIA GPU Operator for automated GPU driver and device setup.[18]
• AMD Spur / Spur-Cloud: A newer GPU-first scheduler with Slurm-compatible CLI and APIs, dashboard-based GPU session launch, and support for bare metal or VMs; it is not Slurm itself, but it targets a similar self-service user experience.[14]

For bare-metal GPU provisioning specifically, the clearest fit in your results is OpenNebula + NiCo and the bare-metal provisioning stack described by NorthWind/Base Command Manager-style automation in the ecosystem.[5][9]

If you want, I can narrow this to:

• enterprise-managed
• open-source
• true bare-metal only
• Kubernetes-based self-service

Source URLs:

• northwind.co
• northwind.co
• opennebula.io
• developer.nvidia.com
• rocm.blogs.amd.com