Questions and answers from Best Practices on NRIS Clusters May 2023, Day 3 (May 11)

Icebreaker questions

  • Which software would you like to run on our HPC machines?

    • COMSOL and Cogenda Genius

    • I hope it is relevant, but I am curious about how to open jupyter notebook, when we are not project leaders.

  • When you need/install new software, is it typically …

    • Python: ooo

    • Conda:

    • R:

    • EasyBuild:

    • Container:

    • C/C++:

    • Fortran:

    • Not sure:

  • Do you want to learn about GPU programming ? Check this next event: https://enccs.se/events/list/?tribe-bar-search=gpu+programming

  • Which GPU programming model do you use?

    • OpenACC:

    • OpenMP offloading

    • SYCL:

    • CUDA:o

    • HIP:

    • OpenCL:

  • Which ML framework do you use?

    • PyTorch:o

    • TensorFlow:

    • Jax:

    • Others:

Questions and Comments

Installing software

https://documentation.sigma2.no/software/modulescheme.html

  1. How to unload a module?

    • module unload [module] Removes just that module

    • module purge Removes all loaded modules

  2. How about mamba? Worthy to make mamba available along with conda?

    • good suggestion! mamba is an alternative implementation to conda which offers sometimes faster dependency resolution

    • RB: personally I use micromamba in a container: https://github.com/bast/singularity-conda

  3. How to do “conda init” without modifying .bashrc

    • source $EBROOTMINICONDA3/bin/activate

    • how is it for anaconda?

      • source $EBROOTANACONDA3/bin/activate

  4. How about this path Source the conda environment setup The variable ${EBROOTANACONDA3} or ${EBROOTMINICONDA3}

source ${EBROOTANACONDA3}/etc/profile.d/conda.sh
source ${EBROOTMINICONDA3}/etc/profile.d/conda.sh

- is this in the documentation? is this a question or an advice? :-)

Question?
Source: https://www.uio.no/english/services/it/research/platforms/edu-research/help/fox/installing-software-python.md
 - see above: `source ${EBROOTANACONDA3}/bin/activate` (or replace by miniconda)
 - it could be that the uio suggestion also works

  1. How about if we want to use a jupyter notebook that other person created, when we are working at the same project?

    • is the notebook then saved on the cluster or where is it saved/shared? -In the cluster/projects/

      • probably NIRD toolkit is a good solution for this

  2. Suggestion on how to run jupyter notebook in compute nodes having GPU as GPU is not available in login nodes?

    • probably NIRD toolkit is a good solution for this

  3. Hm I remember that they told me to use NIRD Toolkit the project leader should do this and give afterwards permission to other users. Might i misunderstand.

    • the project leader should apply for NIRD toolkit to be enabled on a NIRD project to sigma2, and add the people that should be allowed to start a jupyter notebook to a feide group that will be connected to the project. This is a one time thing. After that the project leader does not need to do anything else.

  4. This ticket may be relevant for this discussion

    • The user has wanted to shorten the long custom path

    • https://docs.conda.io/projects/conda/en/latest/user-guide/tasks/manage-environments.html#specifying-a-location-for-an-environment

      • This explains how you can set the env_promt to whatever you want (this case, remove absolute path)

  5. This question may not be relevant: if I need HPC access for 3 years project, how should I apply for the access. Should I apply for SSEW and renew it many times?

    • you can apply for a regular HPC project here: https://www.sigma2.no/high-performance-computing . The application needs to be renewed two times a year. The allocation period is half a year at the time.

    • adding to this: the renewal is typically relatively easy, you don’t need to re-write everything every time. the motivation why quota is allocated on a 6-month basis and not for next 3 years is that often applications are not very good at estimating what they will need (some need a lot more, some use only 10% of what they ask for) and also this gives more flexibility to adapt to new hardware procurements.

    • all HPC questions are relevant and welcome :-)

GPUs

Our general contact email is support@nris.no