Advanced Topics =========================== Here you can find a list of advaced usage instruction for job submission. Container Runtime ----------------- We support the submission of Slurm jobs in OCI compliant runtimes. The **Apptainer container system** (formerly `Singularity`) is available on the compute nodes for both interactive and non-interactive jobs. To learn more about Apptainer, see the `official documentation `__. You can run containers using the Apptainer CLI commands. Alternatively, at submission, you can specify the path of your container using the ``--container`` option. This option is valid for both the ``sbatch`` and ``srun`` command. The value of the ``--container`` option should be a **fully-qualified path**. For example, if the container ``torch_py313.sif`` is stored in your home directory, the ``sbatch`` command should be:: sbatch --container=/home/users/$USER/torch_py313.sif [other options ...] my_script.bash .. Note:: Due to limitations of the GPFS file system, saving unpacked images (`sandboxes`) in your home directory is not supported. Node Constraints ---------------- Slurm allows to constrain node allocation based on pre-defined **node features**. Instead of specifying the partition, request specific features using the ``--constraint`` option. This option is valid for both the ``sbatch`` and ``srun`` command. The value of the ``--contraint`` option should be a string or a comma-separated list of strings. For example, to request nodes that have AMD CPUs connected via NDR InfiniBand:: sbatch --constraint=CPU_MNF:AMD,IB:NDR [other options ...] my_script.bash See `HPC Clusters <../resources/hpc_clusters.html>`__ to check the available features of Magellano clusters.