Snellius provides several compute partitions, each optimised for different types of workloads such as
- CPU-intensive jobs (rome and genoa partitions)
- GPU workloads (gpu_a100 and gpu_h100 partitions)
- large-memory applications (fat_rome, fat_genoa, himem_4tb and himem_8tb), or,
- short interactive tasks (staging, cbuild, gpu_vis)
Users select a partition when submitting jobs through the Slurm scheduler, and the available resources and limits depend on the account / project.
This setup helps balance system usage and ensures that different research workloads can run efficiently on the shared infrastructure.
Usage on Snellius is tracked through an accounting system based on System Billing Units (SBUs).
The number of SBUs consumed depends on the type and amount of resources requested, such as CPUs, GPUs, memory, and runtime.
Different partitions therefore have different accounting weights, reflecting their relative cost and demand. Users and projects can monitor their SBU consumption to manage allocated compute budgets effectively.