funcscan icon indicating copy to clipboard operation
funcscan copied to clipboard

Adapt deepBGC to use GPUs

Open Darcy220606 opened this issue 1 year ago • 4 comments

Description of feature

Adapt deepBGC to use GPU, at the moment it uses CPU and for large input datasets takes ages to run.

Darcy220606 avatar Dec 19 '23 14:12 Darcy220606

I don't think we will be able to do this for 1.2, because of having to re-building the container or something along those lines? @Darcy220606 thoughts?

jfy133 avatar Feb 09 '24 11:02 jfy133

@jfy133 i was told once that we can assign a single process in the pipeline to use GPUs, in the configuration file somewhere here

process {
  withLabel:process_low {
    cpus = { check_max( 2 * task.attempt, 'cpus' ) }
    memory = { check_max( 14.GB * task.attempt, 'memory' ) }
    time = { check_max( 6.h * task.attempt, 'time' ) }
  }

Darcy220606 avatar Feb 09 '24 13:02 Darcy220606

Hmm. Ok.. but I thought the tool has to be compiled for the particular GPU yo be used? I'm speaking completely naively though... Never had to dealt with it before myself

jfy133 avatar Feb 09 '24 16:02 jfy133

Looking at deepBGC repo maybe I'm talking out of my butt...

I have a feeling though, if you are right it's about specifying this in a config file, that would imply you just send it to a node with a GPU? This would then be specific to particular clusters and thus not generally applicable within the code itself? 🤔

/Stream of consciousness

jfy133 avatar Feb 09 '24 17:02 jfy133