- Assign individual GPU IDs to
docker run? The process only uses one GPU but the server may have multiple GPUs, how does each sample know which GPU ID to use?- Is there an established way to detect CUDA versions and choose the right image? I wrote a pipeline to build and tag for every possible version but not sure how to make it happen on nextflow. https://hub.docker.com/r/l1drm/dreg/
I partially solved (2) by a hack:
https://github.com/Functional-Genomics-Lab/dREG-nf/blob/main/modules/local/dreg/main.nf
https://github.com/Functional-Genomics-Lab/dREG-nf/blob/main/bin/cuda_version.sh
