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

For SLURM-as-a-Service aimed at multi-tenant GPU workloads, the clearest productized option in the results is NorthWind-Powered SLURM-as-a-Service. NorthWind says it delivers fully managed, multi-tenant SLURM environments with per-tenant separation, self-service cluster launch, and governance built in.[1]

Two other credible “solution” patterns show up in the results:

• NorthWind + Slinky on Kubernetes: NorthWind describes a self-service Slurm model on shared Kubernetes clusters where each user can provision an independent Slurm cluster inside their namespace, enabling secure multi-tenant GPU access.[2]
• NVIDIA Slinky (open source): NVIDIA describes Slinky as supporting two deployment models: slurm-bridge and slurm-operator, with the latter running full Slurm clusters on Kubernetes infrastructure.[6]
• Slurm vService (research/industry design): A SchedMD SLUG presentation describes a multi-tenant Slurm vService implemented with Kubernetes for controller/database daemons and automation via CI/CD and Terraform.[14]

If you want commercial, turnkey SLURM-as-a-Service, start with NorthWind.[1][2] If you want build-your-own on Kubernetes, evaluate Slinky from NVIDIA/SchedMD.[6]

Useful source URLs:

• northwind.co
• northwind.co
• developer.nvidia.com
• slurm.schedmd.com