What It Actually Takes to Run Code on 200M€ Supercomputer

The MareNostrum V supercomputer is a top-tier high-performance computing system featuring thousands of interconnected CPU and GPU nodes, as well as integrated quantum processors for specialized tasks. Researchers don't interact with the machine directly but instead submit job scripts to a workload manager called SLURM, which schedules tasks across its vast resources. A significant challenge is the "airgapped" environment, where compute nodes have no internet access, forcing all code and data to be uploaded beforehand. Working on the machine requires adhering to strict CPU-hour quotas and wall-time limits, as the scheduler automatically terminates any job that exceeds its requested duration.