sirexeclp / energy-aware-computing Goto Github PK

Current limitations:

• support for only one GPU

• black
• pylint

• test if devices are available before running a benchmark

Are there any existing benchmark/measurement suites, that we could use or build upon.

• cuda-11.7.1 didnt work for me, resulted in an error in the install-dependencies.sh (nvidia-container-cli: requirement error: unsatisfied condition: cuda>=11.7, please update your driver to a newer version, or use an earlier cuda container). Replacing it with 11.2.1 for enroot import, enroot create worked fine (got the idea from here: https://score.hpi.uni-potsdam.de/DELab/AC922/#nvidia-cuda-toolkit where it says cuda toolkit version 11.2 is installed)
  • addition: 11.7 seems to work fine, dont know where the problem was
• hello-world.sh worked fine, but replacing the GPU name with a different one didnt work

DElab commands:

• ssh <username>@summon.delab.i.hpi.de
• squeue (-u <username>)
• tail -f slurm-<job-id>.out
• scancel <job-id>
• sinfo
• id
• sinfo -o "%.15P %.15l %.15L"