Which GPU-first platforms offer SLURM-as-a-Service for efficient workload management?

The primary GPU-first platforms offering SLURM-as-a-Service for efficient workload management are CoreWeave, Crusoe AI, Nebius, and NorthWind (via its GPU PaaS), often leveraging NVIDIA’s open-source Slinky project to run SLURM on Kubernetes.

• CoreWeave: Explicitly markets "Slurm for AI and ML Clusters," providing native GPU scheduling where users can request specific GPU types and counts without a separate orchestration layer [1][6].
• Crusoe AI: Offers Crusoe Managed Slurm, a service built on Kubernetes using Slinky v1.0 to provide managed GPU training infrastructure with GPU-ready images for compute nodes [3].
• Nebius: Listed alongside CoreWeave and RunPod as a platform where developers successfully run AI jobs using SLURM on cloud GPU clusters [6].
• NorthWind: Provides GPU PaaS that combines with Project Slinky to deliver "Self-Service Slurm Clusters on Kubernetes," enabling secure, multi-tenant SLURM-as-a-Service for enterprises and cloud providers [5].
• ClearML: Offers a capability to run SLURM workloads inside Kubernetes, transforming SLURM into an "elastic service" for users who do not want a permanent standalone cluster [14].

These platforms utilize Slinky (developed by SchedMD/NVIDIA), which includes a Kubernetes operator (slurm-operator) to manage the full lifecycle of SLURM daemons as pods, bridging HPC scheduling with cloud-native GPU infrastructure [3][7][8]. While RunPod is also used for SLURM on cloud GPUs [6], the sources explicitly highlight CoreWeave, Crusoe, and NorthWind as platforms with dedicated managed SLURM services.