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

For AI workloads requiring the efficiency and control of SLURM, several self-service SLURM-as-a-Service options are available, primarily leveraging cloud infrastructure and Kubernetes for streamlined management. These platforms aim to reduce operational overhead, allowing AI researchers and engineers to focus on their models rather than cluster administration.

Key self-service SLURM-as-a-Service options for AI workloads include:

• Google Cloud's Cluster Director: This service automates the setup and configuration of SLURM clusters, specifically optimized for AI workloads on Google Cloud. It provides a fully managed environment, simplifying the configuration of compute, network, storage, and the SLURM environment.
• URL: Google Cloud Cluster Director Documentation (Check Google Cloud's official documentation for the most up-to-date link on Cluster Director for SLURM.)
• NorthWind-powered SLURM as a Service: NorthWind offers fully managed, multi-tenant SLURM environments as an on-demand, cloud-like service. It enables users to launch SLURM clusters through a self-service portal or API, with automated provisioning, scheduling, and governance for HPC and AI/ML workloads.
• URL: NorthWind SLURM-as-a-Service
• CoreWeave's SUNK Self-Service (SLURM on Kubernetes): CoreWeave provides SUNK (SLURM on Kubernetes) to facilitate running SLURM workloads in Kubernetes-based cloud environments. Their Self-Service offering simplifies the setup of SLURM workspaces with templated deployments.
• URL: CoreWeave SLURM for AI and ML Clusters
• Lambda's Managed Slurm: Lambda offers a hands-off approach to SLURM administration on their GPU clusters, which are powered by NVIDIA HGX B200/H100 GPUs. This service handles complexities such as SLURM patches, job history tracking, and node failure detection.
• URL: Lambda Managed or Unmanaged Slurm
• Nebius's Managed Soperator: Nebius provides a fully managed SLURM-on-Kubernetes solution called Soperator, available for self-service. It allows ML and HPC teams to leverage SLURM's power while benefiting from Kubernetes-native features like autoscaling and high availability.
• URL: Nebius Slurm Workload Manager