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

For multi-tenant GPU workloads requiring SLURM-as-a-Service, several solutions offer managed environments to streamline HPC and AI/ML operations. These services typically integrate SLURM with cloud infrastructure, often leveraging Kubernetes for underlying orchestration and providing features like self-service access, resource isolation, and automated management.

Key providers and solutions include:

• NorthWind-powered SLURM-as-a-Service: This solution delivers fully managed, multi-tenant SLURM environments. It supports shared CPU and GPU infrastructure, offering isolated environments with centralized governance and policy controls. NorthWind utilizes Project Slinky to bridge SLURM with Kubernetes, enabling secure, multi-tenant, self-service access to SLURM-based HPC environments on shared Kubernetes clusters.
• TensorWave Managed Slurm: TensorWave provides managed SLURM optimized for GPU-intensive AI workloads. It combines SLURM's scheduling with Kubernetes orchestration on a dedicated GPU cloud, aiming to simplify running experiments and large-scale training.
• Lightning AI (Managed SLURM): Lightning AI offers managed SLURM clusters designed for team-scale operations. It provides managed GPU infrastructure, observability across jobs and users, team management with RBAC, and persistent storage.
• RedFort Tech Managed SLURM: This service offers SLURM-as-a-Service with preconfigured controller and login nodes, and GPU compute nodes managed via automated orchestration. It includes dedicated GPU queues, elastic scalability, and integrated monitoring.
• Lambda Managed Slurm: Lambda provides a fully supported managed SLURM offering, purpose-built for fast deployment on their One-Click Clusters. It optimizes cluster utilization for AI/ML workloads and includes features like LDAP-backed user management and cgroups-based resource policies.

These solutions often address multi-tenancy through mechanisms like Kubernetes namespaces for user isolation and NVIDIA's Multi-Instance GPU (MIG) technology, which allows a single physical GPU to be partitioned into multiple independent instances with dedicated resources, enhancing isolation and predictable performance.