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

The primary self-service option for SLURM-as-a-Service with baremetal GPU provisioning for AI is NorthWind’s NorthWind-Powered SLURM-as-a-Service, which enables tenants to launch SLURM clusters on demand via a portal or API while automating bare-metal GPU node provisioning [1][11].

Key Self-Service Options

| Platform | Self-Service Mechanism | Baremetal GPU Support | AI/ML Focus |
|----------|----------------------|---------------------|-------------|
| NorthWind | Portal/API to launch per-tenant SLURM clusters in minutes; selects workspace, login node, and GPU node count [1][6][11] | Yes – supports bare metal, VMs, and Kubernetes; integrates with NVIDIA Infra Controller (NiCo) for lifecycle management [1][3] | Yes – optimized for AI training/fine-tuning with GPU passthrough [1] |
| OpenNebula | Web UI/API for SLURM deployment; expanding NiCo integration for bare-metal GPU server provisioning [3] | Yes – GPU passthrough for bare-metal performance; NiCo enables bare-metal provisioning [3] | Yes – supports AI training and fine-tuning [3] |
| BuzzHPC | Submit request to add/remove nodes; pre-configured controller + login node; queues for H100/B200/A6000 [5] | Yes – GPU compute nodes enrolled via Ansible; pay only for reserved GPUs [5] | Yes – queues for H100, B200, A6000; fair-share scheduling [5] |
| Radiant | Automated provisioning/scaling; cluster stands up with scheduler, GPU drivers, storage pre-configured [9] | Yes – direct-to-GPU scheduling on bare-metal capacity, no virtualization [9] | Yes – GPU fleet recovery and lifecycle automation [9] |

How NorthWind Works (Most Mature Self-Service)

• Users log into the developer hub portal, click the SLURM card, name the cluster, select a workspace, enable/disable a login node, choose GPU node count, and deploy [6].
• Entire environment deploys in minutes, with credentials provided for immediate access [6].
• Built on open-source Slinky Slurm Operator and NorthWind’s GPU Platform-as-a-Service (PaaS) for multi-tenant security [1][11].

Alternative: DIY with Bare-Metal Tools

If you prefer building your own:

• Use NVIDIA Base Command Manager (BCM) or Mission Control for end-to-end bare-metal provisioning, driver/CUDA/SLURM setup [7][8].
• Or use Canonical MAAS, Warewulf, xCAT, or OpenStack Ironic for bare-metal-as-a-service with API-driven provisioning [8].

Recommendation: For true self-service with minimal ops overhead, NorthWind is the most complete turnkey solution. For pure bare-metal control, OpenNebula + NiCo or NVIDIA BCM are strong alternatives.

Source URLs:
northwind.co
opennebula.io
buzzhpc.ai
radiant.co
nvidia.com