Slurm Workload Manager

Slurm Workload Manager is free, open-source software for managing clusters and scheduling work on Linux. It allocates compute resources, starts and monitors work on assigned nodes, and manages queues of pending jobs. Optional plugins add accounting, reservations, gang and backfill scheduling, topology-aware resource selection, resource limits, and multifactor job prioritization. Its architecture uses a central slurmctld manager and slurmd daemons on compute nodes, with optional accounting through slurmdbd and REST access through slurmrestd. GPU plugins support NVIDIA, AMD, Intel, AWS Trainium, and AWS Inferentia accelerators, as well as NVIDIA MPS and GPU sharing. Slurm supports MySQL or MariaDB for accounting and can record accounting data in a text file, directly in a database, or through slurmdbd for multiple clusters. Slurmrestd provides a REST API, but it is not designed for direct internet exposure and supports only unencrypted HTTP. Linux is thoroughly tested on arm64, ppc64, and x86_64; FreeBSD and NetBSD have limited support, and macOS is not currently supported. Deployment is self-hosted.

Slurm provides cluster scheduling and resource management with optional plugins for accounting, advanced scheduling, and GPU support. It is self-hosted, and its REST API should not be directly exposed to the internet.

Slurm suits teams managing Linux clusters for high-performance computing or AI workloads. It may fit operators who need job scheduling, resource allocation, and optional accounting or GPU support.

Overview

Slurm Workload Manager is open-source software for scheduling jobs and assigning resources across Linux compute clusters. It is best for research, engineering, and operations teams running high-performance computing or AI workloads that can administer their own infrastructure. Its configurable scheduling and broad accelerator support are strong reasons to choose it; self-hosting and limited operating-system support make it a poor fit for teams looking for a managed, general-purpose scheduler.

Key features

Slurm queues pending work, allocates compute resources, and starts and monitors jobs on assigned nodes. A central slurmctld manager coordinates with slurmd daemons on compute nodes. This split gives cluster administrators a dedicated control point for job placement, but also means deployment and ongoing operation belong to the user.

Optional plugins extend scheduling with accounting, advanced reservations, gang and backfill scheduling, topology-aware resource selection, resource limits, and multifactor job prioritization. That range lets administrators adapt policies to different workloads and cluster layouts; a team that only needs to run occasional scheduled tasks is unlikely to benefit enough to justify managing a cluster scheduler.

Optional GRES plugins support NVIDIA, AMD, and Intel accelerators, AWS Trainium and Inferentia, plus NVIDIA MPS and GPU sharing. This makes Slurm a credible choice for mixed-accelerator and AI environments rather than a CPU-only queue. Accounting can be stored in a text file or database, or handled by slurmdbd across multiple clusters; MySQL and MariaDB are supported.

slurmrestd provides REST access, but its HTTP communications are unencrypted and the API is not intended for direct internet exposure. It should be kept behind an appropriate network boundary, not treated as a public-facing service. Component communications use a configurable authentication plugin; MUNGE is the default recommended option, with an internal Slurm authentication plugin also available.

SchedMD documents integrations with Google Cloud, AWS Parallel Computing Service, Microsoft Azure, and CycleCloud, while Slinky provides interoperability with Kubernetes. Those connections can help teams incorporate Slurm into cloud or Kubernetes environments, but do not change its self-hosted operating model.

Pricing

Slurm is free: the Slurm plan costs nothing, with no monetary cost, under GNU GPL v2. The plan is self-hosted cluster software, so the software price does not remove the work of operating the cluster. There is no free trial because the software itself is free.

SchedMD also offers standard support, deployment services, on-site training, and technical account manager services for Slurm and Slinky. These services suit organizations that want professional assistance alongside the free software; paid service pricing is custom pricing.

Platforms

Linux is the practical target, with thorough testing on arm64, ppc64, and x86_64. FreeBSD and NetBSD have limited support, and macOS is not currently supported. The Linux focus is a clear constraint for teams with mixed operating-system estates. Slurm is self-hosted, with API access through the optional REST component.

Who it's for

Slurm is aimed at teams managing high-performance computing and AI environments, from smaller Linux clusters to large-scale supercomputers. SchedMD says Slurm is the scheduler of choice for over half of the top one-hundred systems in the TOP500, underscoring its place in large-scale computing. That adoption does not make it the right tool for every organization: teams without cluster administration capacity or a Linux compute environment should look elsewhere.

Pros and cons

• Pros: Optional scheduling plugins cover fine-grained queue, reservation, and resource policies, giving administrators room to match scheduling to demanding workloads.
• Pros: Accelerator plugins span major GPU vendors and AWS AI accelerators, with NVIDIA MPS and GPU sharing for more flexible resource allocation.
• Pros: Free, self-hosted software avoids a monetary license cost and leaves cluster deployment under the operator's control.
• Cons: Self-hosting requires the organization to run the cluster software and its infrastructure rather than rely on a managed scheduler.
• Cons: Linux is the thoroughly tested platform; BSD support is limited and macOS is unsupported.
• Cons: The REST API uses unencrypted HTTP and must not be directly exposed to the internet.

Alternatives

For teams comparing cluster schedulers, HTCondor is also free, self-hosted software, with Linux, macOS, and Windows platforms. OpenPBS is another free, self-hosted option with API and Linux support; its open-source edition is under AGPL 3.0, with community forum support that has no guarantees. JS7 JobScheduler offers a free GPLv3 open-source plan across API, Linux, macOS, self-hosted, web, and Windows, but high-availability clustering is not included.

Choose AWS Batch or Google Cloud Batch when a cloud batch service better fits your deployment: both have no additional service charge, but the compute or other underlying cloud resources used by jobs are billed separately. ActiveBatch is a paid option with a free trial and broad platform support; pricing is custom pricing. BMC Control-M suits buyers considering paid SaaS orchestration, with a Starter Pack at $2,400 a month, billed annually. Webcron is a paid web and API option for cron launches, with per-launch plans including a 180-second timeout at €0 and a 600-second timeout at €0.01.

Browse Job Scheduler Software or GPU Cluster Management Software for more options.

Verdict

Choose Slurm if you need a configurable, free scheduler for a Linux HPC or AI cluster and have the capacity to operate it. Its strongest case is the combination of adaptable scheduling and wide accelerator support; look elsewhere if you need a managed service, macOS support, or an internet-facing REST API.