SCRP compute nodes have Singularity 3.9 installed.


Singularity allows user to run a contained environment potentially different from the host. You can run programs not installed on SCRP or even run a completely different version of Linux.

To use Singularity you need a Singularity image that suits your needs. You can build one from scratch, but it is typically much more convenient to use existing images built by others. You can find images on Singularity Library. Docker containers can also be converted to use with Singularity.

To run a Singularity image for interactive use, type:

compute singularity shell image_path

Note that image_path can be a URL, in which case the image will first be downloaded.

If you need GPU, type instead:

gpu singularity shell --nv image_path

For batch execution, use singularity exec instead of singularity shell.

Both compute and gpu allocate four logical CPUs on a compute node, with the addition of a GPU for the latter. Use srun or sbatch if you need more control over resource allocation.