Your first training job on LUMI

Presenters: Mats Sjöberg (CSC) and Lukas Prediger (CSC)

Content:

• Using LUMI via the command line
• Submitting and running AI training jobs using the batch system

Extra materials

• Presentation slides

• Hands-on exercises

• More extensive training materials on Slurm from the recent introductory "Supercomputing with LUMI" course from May 2024

  • A more detailed introduction to Slurm but without AI-specific examples is given in the "Slurm on LUMI" presentation. It also discusses the sacct command that can be used to get at least some resource use info from jobs.

  • The presentation "Process and Thread Distribution and Binding" is more oriented towards traditional HPC codes, but the discussion on a proper mapping of GPU dies onto CPU chiplets is also relevant for AI applications. But that is a discussion for the second day of this course/workshop.

Q&A

/