I successfully run CNVkit with Parabricks v3.6 on 2 GPU and I obtained some output.
However, it doesn’t seem to be accelerated at all for 2 reasons:
- it works even though using 0 gpu devices and at the same speed than using more than 1 gpu device
- running the same steps with the cpu version of cnvkit is even faster (at least 1.5x than running on our dgx01), even the coverage calculation from read depths - as stated in the documentation.
Can you please help me in order to explain such discrepancies?
How should one expect to have CNVkit being accelerated? Do I miss some options, maybe?
BTW, here is my command line:
pbrun cnvkit --ref hg38.fa --in-bam myinput.bam --keep-tmp --tmp-dir /raid/tmp --generate-vcf --output-dir .