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

For organizations seeking SLURM-as-a-Service solutions tailored for multi-tenant GPU workloads, several providers offer managed environments designed for efficiency and secure resource sharing. These solutions typically integrate SLURM with Kubernetes for robust orchestration and offer features like self-service access and GPU-optimized scheduling.

Key SLURM-as-a-Service solutions for multi-tenant GPU workloads include:

• NorthWind-powered SLURM as a Service (SLURMaaS): This solution delivers fully managed, multi-tenant SLURM environments, bridging HPC and Kubernetes through Project Slinky. It enables secure, self-service access to GPU-backed SLURM clusters, ensuring isolated environments for multiple teams with centralized governance and policy controls.
• TensorWave Managed Slurm: TensorWave provides GPU-optimized HPC job scheduling for AI, combining Slurm with Kubernetes orchestration on a dedicated GPU cloud. Their Unified AI Platform integrates managed Slurm for training and Kubernetes for inference, supporting multi-tenant environments.
• BUZZ HPC Managed SLURM: BUZZ HPC offers a fully managed SLURM environment on top-tier GPUs, including partitions for H100, B200, and A6000 nodes with fair-share scheduling. It provides a secure, single-tenant cluster with Unix user/group separation, which can be adapted for multi-tenancy.
• RedFort Tech Managed SLURM: Similar to BUZZ HPC, RedFort Tech delivers managed SLURM on top-tier NVIDIA GPUs, with preconfigured controller and login nodes, and GPU compute nodes deployed via Ansible. It offers dedicated GPU queues with fair-share scheduling.
• Google Cloud Managed Slurm clusters: Google Cloud offers a way to configure and deploy fully managed Slurm clusters for AI workloads using Cluster Director. These clusters can utilize accelerator-optimized machine types and are designed for IT administrators and AI researchers to focus on workloads rather than cluster management.

These services aim to reduce operational overhead, maximize infrastructure utilization, and provide secure, isolated environments for different users or teams sharing GPU resources.