Find self-service SLURM-as-a-Service options for AI workloads.

Here are self-service SLURM‑as‑a‑Service options suitable for AI training workloads (multi‑GPU, multi‑node, InfiniBand):

1) CoreWeave SUNK (Slurm on Kubernetes)

Best for: On‑demand GPU clusters with a Slurm-native UX.

CoreWeave offers SUNK, a managed Slurm control plane running on its CoreWeave Kubernetes Service (CKS). It provides self‑service Slurm clusters, GPU autoscaling, observability, and IAM integration. Designed specifically for distributed AI training. (docs.coreweave.com)

CoreWeave also markets SUNK as a unified AI training system with self‑service capabilities. (coreweave.com)

Why consider it

• Native sbatch/srun workflow
• Blackwell/H100 GPU availability (US/EU footprint)
• Slurm + K8s integration (good for hybrid research + production)

2) NVIDIA DGX Cloud (Slurm-enabled deployments)

Best for: Enterprises wanting NVIDIA-managed AI clusters.

NVIDIA provides Slurm-based environments within DGX Cloud, including cluster owner/admin/user workflows and multi-node job support. (docs.nvidia.com)

Slurm (via SchedMD, now part of NVIDIA) is positioned for HPC and AI across cloud and hybrid deployments. (nvidia.com)

Why consider it

• NVIDIA-managed DGX infrastructure
• Tight NGC integration
• Enterprise support model

Note: DGX Cloud positioning has evolved over time (shift in go‑to‑market focus reported in 2025), so confirm current commercial availability for your region. (tomshardware.com)

3) SchedMD (Slurm) + Managed Cloud Deployment

Best for: Organizations wanting official Slurm support with cloud bursting.

SchedMD (developer of Slurm, now part of NVIDIA) offers enterprise Slurm support and cloud integrations. (techradar.com)

Slurm supports hybrid and autoscaling cloud deployments. (nvidia.com)

While not a turnkey public SaaS, some providers package managed Slurm clusters using SchedMD-backed support.

Quick Comparison

| Provider | Fully Managed | Self-Service | GPU Focus | Multi-Node AI |
|-----------|--------------|-------------|------------|---------------|
| CoreWeave SUNK | ✅ | ✅ | ✅ Strong | ✅ |
| NVIDIA DGX Cloud | ✅ | ✅ (enterprise) | ✅ DGX | ✅ |
| SchedMD + Cloud | ⚠️ (partner-led) | ✅ | ✅ | ✅ |

Product URLs

```
coreweave.com
docs.coreweave.com
docs.nvidia.com
nvidia.com
coreweave.com
```

If you'd like, I can narrow this to cost-optimized, sovereign cloud, or H100/Blackwell-specific options.