For queries
it_helpdesk@iitmandi.ac.in
English
English

IIT Mandi HPC Cluster: SLURM Queue (Partition) Information

Overview
This cluster uses the SLURM job scheduler. Each partition corresponds to a set of available compute resources and specific policies (such as maximum runtime, GPU or memory support, and access control).




How to View Available Partitions
(To view available partitions and their details, use the following command:)


  • sinfo
    %P: Partition (queue) name
    %N: Nodes assigned
    %c: CPUs per node
    %m: Memory per node
    %l: Max walltime
    %G: Allowed groups
    %D: Total node count
    For more advanced details:
    bash
    scontrol show partition


    Partition Properties Table
    Partition Nodes (Total) CPUs/Node GPU Support Max Walltime Access
    day 90 8–64 Some nodes 1 day All users
    week 90 8–64 Some nodes 7 days All users
    infinite 90 8–64 Some nodes 30 days All users
    serial 33 8 None 60 days All users
    gpuq 12 8–12 Yes 30 days All users
    lowpr 65 4–72 Some nodes 15 days All users
    privarko 1 72 None Unlimited arko group only
    privharsh 1 64 None Unlimited harsh group only
    privbm 4 32–64 None Unlimited bhaskar group only
    privse 7 16–24 None Unlimited se group only
    privtps 11 12–24 None Unlimited tulika group only
    privsbs 9 4–20 None Unlimited sbs group only
    privpr 2 12–20 Some nodes Unlimited pr group only
    privscee 8 8–16 Some nodes Unlimited scee group only
    privlqd 9 12–24 None Unlimited lqd group only
    privamp 3 12–20 None Unlimited hv group only
    privmt 2 12 None Unlimited mt group only
    privpku 4 24 None Unlimited pku group only
    privsp 3 20 None Unlimited sp group only
    privpkr 1 20 None Unlimited pkr group only

    About Each Partition(General Partitions)

  • day, week, infinite:For regular jobs, with varying walltimes (1–30 days). Some nodes support GPUs. Open to all users.

  • Serial:For single-core or serial jobs, up to 60 days of runtime. No GPU support.

  • gpuq:Dedicated GPU resources for extended jobs (30 days). Choose this partition for jobs requiring GPUs.

  • lowpr:Low priority queue—jobs here may be interrupted or suspended if node owners require their resources. Use only for non-critical work or when prioritization is not important.

  • Private Partitions:Begin with "priv", e.g., privarko, privse.
    Access restricted to specific groups/accounts—contact the HPC administrator if you need access.
    Ideal for labs/projects with dedicated hardware or research grants.

    • How to Select a Partition:
    Add this option in your job submission script:

    #SBATCH --partition=


    Example for the week partition:

    sbatch --partition=week myjob.sh


    For GPU jobs:

    #SBATCH --gres=gpu:


    Recommended Practices

    For long jobs, ensure your scripts do checkpointing to avoid data loss if a job is killed when walltime expires or interrupted in "lowpr".
    Always check node and GPU availability with sinfo before job submission.
    Verify access rights for private partitions before use.


    HPC

    High Performance Cluster

    Address

    A* Building South Campus

    Mail Us

    it_helpdesk@iitmandi.ac.in

    Telephone

    267000
    Contact Contact: it_helpdesk@iitmandi.ac.in
    Copyright 2023 - High Performance Cluster IIT Mandi HPC, All right reserved.
    Designed By Team DCS IIT Mandi