Closed dkainer closed 5 years ago
@dkainer --numThreads
is not in the manuals, is it implemented?
@dkainer numThread helps in certain analysis scenario. It helps most when there are large matrix computations. So 16 threads can be slower than 1 threads in certain cases. What is your use case?
@zx8754 numThreads uses OpenMP. @dkainer it does not use MPI.
@zhanxw Thank you. Just wanted to point out that I couldn't see that option (--numThreads
) in the manuals. But we have --thread N
option, is it the same?
The option --thread N
is for vcf2kinship, but --numThread N
is for rvtest. Sorry for the inconsistency.
--thread N
is documented in the manuals (http://zhanxw.github.io/rvtests/) as below:
When dealing with large input files, it is often preferred to use multiple CPU to speed up calculation using the option --thread N in which N is the number of CPU.
But --numThread N
is not in manuals. Could this be added to manuals?
Yes, I just did. Thanks.
I am running rvtest on a single node of a cluster. It works, but for some reason if I set --numThreads 16 it runs much slower than when using just --numThreads 1
With --numThreads 1 it completed 95,000 regions for one phenotype in about 140 mins With --numThreads 16 it completed 95,000 regions for one phenotype in about 570 mins
My command to run rvtest is prefaced with the 'aprun' MPI command so perhaps I need to compile rvtest with some special flags to use the multiple threads properly?
best David