Users and Services Portal - Help & Docs

# Partitions queues

In the SLURM scheduling system, partitions (often referred to as queues) define how users’ jobs are submitted, prioritized, and limited in terms of resources and runtime.
Physically, the same compute nodes can belong to multiple partitions, but each partition serves a specific purpose — for example, general workloads, quick tests, long runs, or GPU computing.

# Partition

  • standard
  • fast
  • long
  • tesla

# Partition standard

  • Purpose: Default partition for most computations.

  • Nodes: All standard compute nodes of the cluster.

  • Max runtime: 7 days (168 hours)

  • Use case: Regular production jobs that don’t require special resources.

Notes: Recommended for the majority of users and applications.

# Partition fast

  • Purpose: Short, quick, high-priority jobs (mainly testing or debugging).

  • Nodes: Shared with standard partition.

  • Max runtime: 1 hour

  • Priority: Higher than standard to allow rapid testing turnaround.

  • Use case: Compilation, testing, or short benchmarking.

# Partition interactive

  • Purpose: Interactive, real-time jobs (testing, debugging)
  • Nodes: Standard compute nodes
  • Max runtime: 10 hours
  • Default runtime: 1 hour
  • Priority: Very high
  • Use case: Quick interactive work using

# Partition long

  • Purpose: Extended runtime computations that exceed the standard 7-day limit.

  • Nodes: Selected subset of compute nodes.

  • Max runtime: Up to 30 days (720 hours)

  • Use case: Long-term simulations, parameter sweeps, or heavy workloads requiring long walltime.

  • Notes:

    • Jobs in this queue may start later due to longer scheduling horizon.
    • Access to this partition is granted on request — users must contact the administrator to be added.

# Partition tesla

  • Purpose: GPU-accelerated computations.

  • Nodes: Equipped with NVIDIA Tesla V100 and H100 GPUs

  • Max runtime: 7 days (168 hours)

  • Use case: Deep learning, molecular dynamics, GPU-based simulations.

  • Notes:

    • Requires proper CUDA or GPU-aware software configuration.
    • For instructions on how to submit GPU jobs, see How to submit GPU jobs (opens new window)
    • Please note that not all GPU nodes have identical configurations; select GPU resources accordingly in your job script.