Skip to content
Type to start searching

    Table of Contents

    Available batch job partitions

    Puhti partitions

    The following partitions (aka queues) are currently available in Puhti for normal (CPU) nodes:

    Partition Time
    limit    		 Max
    tasks    		 Max
    nodes    		 Node types Max
    memory    		 Max
    local storage
    (nvme)
    test 15 minutes 80 2 M 190 GiB
    interactive 7 days 4 1 IO 64 GiB 640 GiB
    small 3 days 40 1 M, L, IO 382 GiB 3600 GiB
    large 3 days 1040 26 M, L, IO 382 Gib 3600 GiB
    longrun 14 days 40 1 M, L, IO 382 GiB 3600 GiB
    hugemem 3 days 160 4 XL, BM 1534 GiB
    hugemem_longrun 7 days 40 1 XL, BM 1534 GiB

    The following partitions are available on GPU nodes. Note that for each GPU, you should reserve at most 10 cores/task.

    Partition Time
    limit    		 Max
    GPUs    		 Max
    nodes    		 Node types Max
    memory    		 Max
    local storage (nvme)
    gputest 15 minutes 8 2 GPU 382 GiB 3600 GiB
    gpu 3 days 80 20 GPU 382 GiB 3600 GiB

    Information about the partition:

    sinfo

    or 

    scontrol show partition <partition_name>

    Notes on partitions

    • Only request the memory you need
      • The jobs will run sooner, since in each partition, most of the nodes have less memory than the maximum amount
      • Billing is based on memory requested, not on memory used
    • Only use longrun partitions if necessary
      • These partitions have stricter limits and a lower priority
    • Only one job per user is allowed in interactive partition

    Mahti partitions

    The following partitions (aka queues) are currently available in Mahti on CPU nodes.

    Partition Nodes Time
    limit    		 Access
    test 1-2 1 hours all
    medium 1-20 36 hours all
    large 20-200 36 hours scalability test
    gc 200-700 36 hours Grand Challenge
    interactive 1 7 days all

    The following partitions are available on GPU nodes.

    Partition Time
    limit    		 Max
    GPUs    		 Max
    nodes    		 Node types Max
    local storage (nvme)
    gputest 15 minutes 4 1 GPU 3800 GiB
    gpusmall 36 hours 2 1 GPU 3800 GiB
    gpumedium 36 hours 24 6 GPU 3800 GiB

    Interactive partition, is intended for pre- and post processing tasks. It can be used both for interactive working and in batch jobs, where reserving a full node is not reasonable.

    The basic CPU based HPC computing should be done in medium, large or gc partitions. Only full nodes are allocated from these partitions and the jobs will have access to all cores and memory in the nodes. You may choose to run with fewer cores if that gives better performance. In these partitions billing is based on the allocated nodes.

    The large partition on Mahti is only accessible to projects that have done a scalability test and shown good utilisation of the resources. The partition gc is only accessible to Grand Challenge projects and allows users to run extremely large simulations.

    Note that for each GPU, you can reserve at most 32 cores/task and for each GPU is allocated 122.5 GiB of memory.