Thanks David,
OMP_NUM_THREADS did the trick. I experimented with different values and in
the end, just OMP_NUM_THREADS=1 was the best option, at least for ek
default values used by sqm under antechamber.
On Fri, 31 Dec 2021 at 14:08, David A Case <david.case.rutgers.edu> wrote:
> On Fri, Dec 31, 2021, Alan wrote:
> >Hi there,
> >
> >So I went to test an old molecule I know well and put it to run under sqm
> >(from AT 21 via Conda) in two different machines:
> >
> >1. Old mac cpu Intel i5 dual-core 2.6 GHz
> >2. Linux server 36 cpus . Intel(R) Xeon(R) CPU E5-2697 v4 @ 2.30GHz
> >
> >First, to my surprise, I saw 36 sqm threads running. I didn't know it was
> >capable of running in parallel!
>
> There is a short section on "parallel simulations" in Section 11.1.5, but
> it
> is mostly about MPI, and only the final sentence is likely to be helpful to
> you.
>
> Two things to play with: the environment variable OMP_NUM_THREADS, which
> determines the OpenMP thread count, and the "diag_routine" variable that
> controls how matrix diagonalizations are done. The optimal number of
> threads is almost certainly far smaller than 36.
>
> Others on the list are much more familiar with the innards of sqm than I
> am,
> and might have suggestions for parallelization.
>
> ....dac
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
Alan Silva 🚲🏊♂🏃♂🇧🇷🇫🇷🇬🇧
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sat Jan 01 2022 - 03:00:02 PST
