Dear Amber user:
I successfully used the gfn-xTB in amber QM/MM, by modifying sander/qm2_extern_orc_module.F90, but low parallel efficiency problem appeared.
However both 'num_threads=12' in orc namelist of amber input file 、'%pal NPROCS 12 end' in orc_job.tpl and 'export OMP/MVK_NUM_THREADS' can not be efficiently parallel.
Both scheme can only increase the speed by up to about 10% than 1 core scheme, and even reduce the computational efficiency, I think this is abnormal. Can someone give me any help here?
#SI
software: Amber18(gnu openmpi) openmpi 3.1.4 orca 4.2.1 gcc 4.8.5 openMP 201107
test reslut
method orc-num_threadsxtb-pal NPORCS OPENMP ps/d
xtb 1 1 164
xtb 1 1 1267
xtb 12 1 163
xtb 1 12 121.8
b3lyp 1 1 /0.17
b3lyp 20 1 /0.17
b3lyp 1 20 / 1.83
Regards,
Jia-Ning Wang
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Mar 10 2021 - 20:30:02 PST
