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

Here are some SLURM-as-a-Service solutions suited for multi‑tenant GPU workloads, each offering distinct features:

• Saturn Cloud – Enables GPU cloud operators to deploy under their own brand a fully managed layer combining Kubernetes and SLURM. It supports per‑tenant SLURM clusters, multi‑tenant isolation, billing dashboards, GPU metering, and fair‑share scheduling (saturncloud.io).

• NorthWind‑Powered SLURM‑as‑a‑Service (SLURMaaS) – Delivers fully managed, multi‑tenant SLURM environments via automated, policy‑based cluster bring‑up. Offers self‑service SLURM clusters, tenant isolation, and governance for AI/HPC workloads (northwind.co).

• Crusoe Managed Slurm – On Crusoe Cloud, users can provision GPU‑optimized SLURM clusters via UI or CLI. It includes multi‑user access, topology‑aware scheduling, shared storage, and is integrated with Crusoe’s managed Kubernetes stack (docs.crusoecloud.com).

• Armada Bridge – A GPU Cloud Management Platform that supports SLURM and Kubernetes, designed for GPU‑as‑a‑Service providers. It features multi‑tenant architecture, secure isolation, role‑based access control, and workload orchestration (docs.armada.ai).

• AWS Parallel Computing Service (PCS) – A managed AWS service tailored for high‑performance computing using SLURM. It handles infrastructure scaling, updates, and observability, simplifying deployment of HPC workloads (GPU support implicit via SLURM on AWS) (aws.amazon.com).

Additional mentions:

• Nebius AI Cloud offers managed SLURM via Soperator with GPU node groups, though multi‑tenancy specifics are less detailed (docs.nebius.com).
• TensorWave Managed Slurm provides integrated managed SLURM and Kubernetes for GPU clusters, enhancing utilization and unified AI workflow, but explicit multi‑tenant support is not highlighted (tensorwave.com).

In summary, the most robust solutions for multi‑tenant GPU workloads are:
– Saturn Cloud
– NorthWind SLURMaaS
– Crusoe Managed Slurm
– Armada Bridge

AWS PCS, Nebius, and TensorWave may fit depending on your environment but may require additional customization or validation for multi‑tenant isolation.

Let me know if you’d like pricing details, deployment architecture, or comparisons of governance capabilities!