Job script generator for Slurm

This page generates job scripts for Spartan, which uses the Slurm workload manager. It isn't intended to cover every option available, just a little help to get you started with some of the possibilities and how the syntax works.

Enter the details of your job in the form below and then click on the make script button below.

This utility is adapted from work by Bernie Pope, Ben Moran and Lev Lafayette.

Job Name:

It is a good idea to give your job a meaningful name. This will help you identify it when listing the job queue.

Project ID:

Specify the project ID to which this job should be attributed. For example, punim0001.

Partition: sapphire cascade gpu-a100 gpu-h100

Job Type: Single Core Multithreaded (SMP) MPI, single or multiple node

Single Core

Suitable for single CPU core jobs only (not parallel).

Multithreaded (SMP)

Suitable for shared-memory multithreaded jobs. Note these jobs cannot use more than one node and cannot use more cores than the number of cores on a node.

MPI

Suitable for distributed parallel jobs that use MPI. May use more than one cluster node. These are allocated to the "cascade" or "sapphire" partition.

Notes

If the job submission is more complex that these deliberately restrictive criteria (e.g., SMP jobs with more than 12 cores, MPI jobs with specific node allocations, GPU partition and GRES requirements), then you should be at the stage of writing your own scripts!

Close

Specify whether your job needs one or more cores (more...)

CPU Cores (total):

The maximum number of CPU cores needed by your job.

GPUs (per node):

The number of GPUs you would like, per node. If you selected the partition gpu-a100 or gpu-h100, please ensure this value is more than 0

Slurm QoS:

The Slurm QoS for the job.

Required Memory (GB):

Single Core and MPI

For Single Core and MPI jobs this specifies the amount of memory allocated to each CPU core requested by the job. The total memory used is equal to the number of cores in the job multiplied by the amount of memory requested. If you leave this blank then your job will be allocated 4000 megabytes per CPU core.

For example if you request 2 CPU cores and 8GB of memory per core, the total memory used by the job is 16GB.

Multithreaded (SMP) jobs

For Multithreaded (SMP) jobs this is the total memory allocated per node, irrespective of the number of CPU cores.

Close

Leave this blank to use the default (more...)

Notification Email: Job Starts
Job Ends (Success)
Job Ends (Error)

You can be notified by email when your job starts running or when it ends (either successfully or with an error).

Walltime: days hours minutes

Maximum amount of time needed for your job to complete (it will be terminated after this time).

Directory to run the job from:

The same directory where it is launched

Home directory

Other

Each job is run from a given directory on the computer's filesystem - this is called the working directory in Unix terminology.

You need to set the working directory correctly so that your job can find its input files and generate its output files in the appropriate location. In most cases it is desirable to set this to the directory where the job was launched, but you might also like it to be your home directory, or from some other specific directory on the computer.

If you specify the directory, it must be an absolute reference (i.e. /home/foo/mydir) or relative to the launch directory.

Close

The working directory for your job (more...)

Modules:

Single Core and MPI

The modules utility sets up your environment paths for particular versions of specified programs. It is possible to use more than one module in your job (just list all the ones you need on separate lines).

For example to use version 3.2 of the application foo, which was compiled with gcc version 4.9.2, you should load the module called: foo/3.2-gcc-4.9.2

You can list all available modules on Spartan using the module avail command.

Close

List of modules to load for your job, one per line (more...)

Command:

Enter the command (or commands) that you want to run, one per line. Note that MPI jobs need to prefix the command with srun.

Download this script