Hi, Is the

MMSEQS_COLABFOLDSEARCH

process supposed to use the GPU when

--use_gpu true

? I am trying to run Boltz mode without an MSA server on a multimer complex and it goes into running mmseqs2 in CPU mode. My command line is

nextflow run ../proteinfold-nf/dev --mode boltz --input samplesheet-complex.csv \
   --boltz_db ../DB/boltz --colabfold_db ../DB/boltz --outdir t5 \
   --use_gpu true --use_msa_server false \
   -profile apptainer -c custom.config

custom.config

only contains my apptainer cacheDir and

apptainer.envWhitelist = "CUDA_VISIBLE_DEVICES,NVIDIA_VISIBLE_DEVICES"

I set

export CUDA_VISIBLE_DEVICES=1

before the run. The samplesheet contains a single chain (< 500 AA residues all together) in a single fasta file. Needless to say, CPU MSA computation completes without problems in my machine. I can see the apptainer invocation in

MMSEQS_COLABFOLDSEARCH

process passing the

--nv

to the image, but nothing else that indicates it will (or can not) use the GPU in the logs. In my tests I have a small T1000, but even if there is not enough VRAM I would expect some error message. Any clues or words of wisdom ? Thanks.