The HPE Cray Programming Environment

Presenter: Kurt Lust (LUST)

As Linux itself is not a complete supercomputer operating system, many components that are essential for the proper functioning of a supercomputer are separate packages (such as the Slurm scheduler discussed later in this course) or part of programming environments. It is important to understand the consequences of this, even if all you want is to simply run a program.

Materials

• A video recording will follow.

• Slides

• Course notes

• Exercises

Q&A

1. Why is the next system Nvidia ? is it to do with vendor lock

   • LUMI-AI will be an AMD machine (see: https://www.eurohpc-ju.europa.eu/eurohpc-ju-signs-contract-deploy-lumi-ai-supercomputer-2026-08-31_en ).

   • If the confusion comes from the Wave32 model: AMD currently has two architectures: CDNA version 1 till 4 (MI250X is CDNA2) have Wave64, but the some of the RDNA GPUs are also supported by ROCm and they have Wave32. For the next iteration, CDNA5, AMD is moving to a more unified architecture and also goes to Wave32 as the native format for the compute GPUs.