Page tree
Skip to end of metadata
Go to start of metadata
Updated 2020-09-01


Below is a table describing each of the resource partitions available on Viking. These may change as we come to understand the workloads running on Viking and update the partitions to reflect that. 

Partition nameMax job timeMax number of jobs per userMax running cores per userMax running memory per userDefault number of cores per jobDefault memory per coreNumber of nodes in partitionAdditional Information
nodes48 hrsNo Limit5002.4 TB14.8 GB170Default partition for everything - due to the the 500 core per user limit, there is a practical job limit of 500 running jobs, assuming 1 core per job. There is also a limit of 2.4TB in total.
test30 mins

2 running; 4 total

816 GB11 GB170Partition for testing small jobs. Request using --partition=test in job script.
preempt30 daysNo limitNo limitNo limit14.8 GB170

Warning: jobs submitted to this partition are subject to preemption. There is no guarantee when or if a job will run, or whether it will complete.

This partition has no limits on the resources that can be requested, but a lower priority than the nodes partition. Jobs will run if there are free resources in the nodes partition, but will be cancelled if these resources are later required for another higher priority job. Need to request using --partition=preempt in job script.

himem48 hrsNo limit2883.78 TB111.875 GB4

Partition for running jobs that require memory greater than that available in other partitions. Need to request using --partition=himem in job script.

Three nodes (himem02, himem04 and himem05) have 64 cores, one node (himem01) has 96 cores.


The max running cores and max running memory limits are practical limits, due to the resources available on the nodes.

himem_week7 daysNo limit64760 GB111.875 GB1

Partition for running jobs that require memory greater than that available in other partitions. Need to request using
--partition=himem_week in job script.


The max running cores and max running memory limits are practical limits, due to the resources available on the nodes.

gpu48 hrsNo limit40384 GB19.6 GB2

Partition for running jobs that require GPUs, see documentation for details about how to request GPUs.


There are 4 GPUs and 40 cores on each GPU node, subject to the following rules:

  • Your job script must request at least one GPU
  • You can request a maximum of 10 cores per GPU
  • You can use no more than 4 GPUs at a time across all of your jobs running in the GPU partition
week7 daysNo limit4001.92 TB14.75 GB10

Partition for jobs that need to run for longer than the 48h max time allowed by the nodes partition, but for less than a week. Need to request using --partition=week in job script.


The max running cores and max running memory limits are practical limits, due to the resources available on the nodes.

month30 daysNo limit120575 GB14.75 GB3

Partition for jobs that need to run for longer than the 7 day max time allowed by the week partition, but for less than 30 days. Need to request using --partition=month in job script.


The max running cores and max running memory limits are practical limits, due to the resources available on the nodes.

interactive8 hrs11

19258M

19629M1Partition for jobs that would like to use Viking interactively.  This could be used for graphical work with the 8 hr queue limit set to mimic a working day.  Need to request using --partition=interactive.
  • No labels