Control number of CPUs used by deepvariant


I’m currently using deepvariant in the clara parabricks image to perform variant calling. I’m running the command as follows:

pbrun deepvariant \
        --ref $genome \
        --in-bam $bam_file \
        --normalize-reads \
        --num-streams-per-gpu 4 \
        --run-partition \
        --gpu-num-per-partition 1 \
        --gvcf \
        --out-variants ${sample}.${extension} \
        --tmp-dir tmp_dir

For this process, I usually request 24 CPUs and 2 GPUs on one of our GPU nodes equipped with Tesla V100 GPUs. The number of CPUs is set as recommended in Parabricks docs when using 2 GPUs.

However, the IT department reported that the processes are using many more threads than expected on the CPUs, overloading the system.

Is there something I’m missing in the deepvariant configuration, or is there a way to control the number of threads?