[gmx-users] gmx, threads, pinoffset and stride

[Michael Brunsteiner]

doing some test runs to optimize the mdrun settings for my hardwarei noticed a couple of things i fail to understand (everything below is gmx-2018on an intel CPU 6 cores, 2 threads each, and GTX 1060)

1) when i start a run as in, e.g.:  prompt> gmx mdrun -v -nt 12 -ntmpi 1 -ntomp 12  -deffnm mcz >& er &and then look at cpu usage with the command,   prompt> htopthen htop claims that there are not 12 but in fact 14 gmx instances running12 of which use about 100& CPU each, and the remaining two close to 0%such unexpected extra processes i also see with a number of other variations of the gmx mdrun command -is this a problem of htop?
2) at http://manual.gromacs.org/documentation/2018/user-guide/mdrun-performance.htmlit says:
gmx mdrun -nt 6 -pin on -pinoffset 0
gmx mdrun -nt 6 -pin on -pinoffset 3
Starts two mdrun processes, each with six total threads.Threads will have their affinities set to particularlogical cores, beginning from the logical corewith rank 0 or 3, respectively. The above would workwell on an Intel CPU with six physical cores andhyper-threading enabled. Use this kind of setup onlyif restricting mdrun to a subset of cores to share anode with other processes.
... if these are logical and not physical cores (is this threads vs hyper-threads??) shouldn't this actually be:    mdrun -nt 6 -pin on -pinoffset 0
    mdrun -nt 6 -pin on -pinoffset 6 ??
3) can anybody explain what -pinstride really is/does ?is there ever any good reason for this not being 1 ?


cheers,michael




=============================== Why be happy when you could be normal?