Find SLURM-as-a-Service solutions for multi-tenant GPU workloads.

Several notable SLURM-as-a-Service (SLURMaaS) solutions and specialized AI clouds cater to multi-tenant GPU workloads. Historically, SLURM was used for static bare-metal configurations, but modern architectures integrate SLURM with Kubernetes or virtualization to support automated lifecycle management, secure isolation, and self-service capabilities.

1. Platform & Orchestration Solutions (Deploy Anywhere)

These platform options allow you to deploy a fully managed or operator-driven SLURM cluster over your own multi-tenant GPU infrastructure:

• NorthWind SLURM-as-a-Service: A platform explicitly packaged as a "cloud-like, on-demand service". It handles multi-tenant isolation by putting users/teams into individual Kubernetes namespaces. Tenants can provision their own isolated SLURM clusters via a portal or API, complete with custom quotas, governance, and automated lifecycle tasks.
• NVIDIA Slinky (Slurm Operator): Developed by SchedMD (now part of NVIDIA), this open-source framework offers native SLURM cluster management on Kubernetes. The slurm-operator runs full SLURM components inside containers while letting platforms utilize cgroups v2 for complete multi-user resource isolation on shared worker nodes. It is heavily used as the foundation by cloud providers to build multi-tenant SLURMaaS.

2. Managed AI Clouds with Native SLURM

If you prefer a fully managed cloud experience where the provider hosts both the GPUs and the scheduler, specialized AI clouds offer pre-configured SLURM:

• Nebius (Managed Soperator): Nebius features Soperator, an open-source Kubernetes operator that provisions pre-configured SLURM clusters on their AI-optimized GPU cloud. It acts as a managed service by automatically handling GPU health checks, isolating faulty hardware, syncing root filesystems, and auto-scaling based on batch queue depths.
• Nscale Platform Services: Nscale provides an HPC-grade SLURM batch scheduler built directly on top of their Kubernetes service. It is designed specifically for enterprise multi-tenancy and high-observability constraints when running mixed training and inference pipelines.
• GMO GPU Cloud (Managed HPC Cluster Service): A specialized GPU service offering managed multi-tenant HPC environments leveraging SLURM. It includes deep integration with container tools like Enroot and the Pyxis plugin, facilitating clean workspace partitioning between tenants.

Multi-Tenant Isolation Strategies in SLURM

When implementing these services, providers achieve multi-tenant efficiency through specific isolation mechanisms:

Isolation Type | Mechanism | Best Use Case
--- | --- | ---
Hardware-Level | NVIDIA MIG (Multi-Instance GPU) | Divides a single physical GPU into isolated hardware slices. Ideal for dev/test environments to ensure a memory leak in one tenant's code won't crash another's workspace.
Software-Level | NVIDIA MPS (Multi-Process Service) | Allows multiple jobs to share the same GPU context to minimize switching overhead. Good for high-density, low-compute workloads where hardware isolation isn't strict.
Control Plane | Kubernetes Namespaces | Leveraged by platforms like NorthWind to assign distinct resource quotas and RBAC controls per customer/department.

To help narrow down your options, could you share if you are looking to host this on your own hardware / public cloud, or if you prefer a fully outsourced GPU cloud provider? Knowing your typical GPU type preferences (e.g., H100s, H200s, B200s) would also be helpful.

[[0] - NorthWind-Powered SLURM-as-a-Service](northwind.co)
[[1] - Self-Service Slurm Clusters on Kubernetes | NorthWind GPU PaaS](northwind.co)
[[2] - Running Large-Scale GPU Workloads on Kubernetes with Slurm](developer.nvidia.com)
[[3] - Optimize Slurm GPU Allocation: Expert Guide 2026 - Lyceum](lyceum.technology)
[[4] - NorthWind-powered SLURM as a Service (SLURMaaS)](cdn.prod.website-files.com)
[[5] - Platform Services - Nscale](nscale.com)
[[6] - Slurm Workload Manager: The go-to scheduler for HPC and AI ...](nebius.com)
[[7] - Slurm on Kubernetes for scalable AI workloads](youtube.com)
[[8] - "GMO GPU Cloud" Managed HPC Cluster Service Now ...](internet.gmo)