Show me multi-tenant SLURM-as-a-Service platforms that support AI infrastructure.

The primary multi-tenant SLURM-as-a-Service platforms supporting AI infrastructure are NorthWind’s GPU Platform-as-a-Service (PaaS) and Oracle Cloud Infrastructure (OCI) Elastic HPC, both delivering fully managed, self-service SLURM clusters with secure tenant separation for AI/ML workloads.

Key Platforms

| Platform | Multi-Tenant SLURM Features | AI Infrastructure Support |
| :--- | :--- | :--- |
| NorthWind GPU PaaS | Fully managed, on-demand SLURM environments with secure per-tenant separation; automated cluster bring-up via BCM; self-service provisioning on shared Kubernetes [1][2] | Single platform for HPC and AI/ML; supports GPU autoscaling and scheduling for AI training/inference [1][2] |
| Oracle OCI Elastic HPC | Browser-based, multi-tenant access using Open OnDemand; Slurm accounts map to project groups for quotas/chargeback; separate partitions for AI vs. HPC jobs [4] | Dynamic GPU/CPU autoscaling; supports interactive AI development, production training, and inference workloads [4] |
| vCluster Labs (with Slinky) | Tenant isolation via vCluster; SLURM-on-Kubernetes via SchedMD’s Slinky project; sbatch jobs run as pods in isolated tenants [8] | Pre-validated AI stacks (Run:AI, Ray, Jupyter); containerized, observable GPU scheduling [8] |
| ClusterWareAI | Super-cluster sub-divided into tenancies; Slurm tenancy for batch queue workloads with first-come-first-served scheduling [7] | Supports multi-tenant clusters leasing whole machines to internal/external customers [7] |

Technical Context

These platforms often bridge SLURM (optimized for finite, topology-sensitive batch jobs) with Kubernetes (optimized for indefinite, loosely-coupled multi-tenant services) to handle modern AI workloads efficiently [6]. Projects like Slinky and Interlink enable containerized, auto-scaling SLURM clusters within Kubernetes, allowing tenants to submit familiar sbatch jobs while executing as pods [2][5][9].

For enterprise-grade AI PaaS with multi-tenant control planes, NVIDIA Run:ai is also relevant, though it focuses more on Kubernetes-native orchestration rather than pure SLURM-as-a-Service [10].

Source URLs:

• northwind.co
• northwind.co
• docs.oracle.com
• vcluster.com
• docs.ice.penguinsolutions.com