maximizing CPU usage

[Alan-Abdallah]

Hi :)

I want to compile AD-GPU to run (in batch-mode) on a 48-core CPU node (2 x Intel(R) Xeon(R) Gold 6252).
/AD_gpu/ad_gpu -I receptor.reactive_config -D all -H 0 -x 0 -B {filelist} -N {p_output}/ -n {amount_poses}

is there a correct way to balance the Workgroup-Size and the amount of OpenMP-Threads for CPU-only usage?

should I do:
make DEVICE=CPU NUMWI=1 OVERLAP=ON and OMP_NUM_THREADS=48

or something like:
make DEVICE=CPU NUMWI=16 OVERLAP=ON and OMP_NUM_THREADS=3

[atillack]

Hi @Alan-Abdallah, I don't know what the best approach here would be for sure but intuitively I would give OpenCL a full CPU each - so either NUMWI=24 or NUMWI=48 (# of threads per 6252).

I never tested this configuration though - as our servers don't have Intel's OpenCL icd, so I would recommend to first optimize the runtime of a single test ligand (e.g. make test) to get an optimal NUMWI (i.e. the one with the best performance per NUMWI) and then go from there.