On Mon, May 9, 2011 at 4:52 PM, Guenegou, Guillaume [ORDFR] <
GGUENEGO.its.jnj.com> wrote:
> Dear all,
>
> As some of you might know, I am trying to switch from amber 8 to amber
> 11.
>
> Regarding some calculations durations, it seems sander is slower with
> amber 11 than with amber 8. We will all agree it is an unexpected
> result.
>
Actually I'll disagree here. If you want to test performance, what you
should test is pmemd8 vs pmemd11. pmemd is the performance engine (and
you'll find that it's much faster than in amber8). sander is a general
purpose code that is modified by all developers that implement new ideas.
As such, the focus of sander development is implementing new features, NOT
performance, and the core routines that take a long time (i.e. non-bonded
calculations and the FFT in PME) probably haven't undergone any optimization
since amber8, so there's little to no chance that the code would actually
get *faster*.
In fact, if anyone implements any new feature in these loops (for instance,
constant pH or softcore TI which were implemented *after* amber8, as I
recall), you're introducing conditionals into the inner loops of the
non-bonded calculations, which will add some overhead that wasn't present
before, causing slowdowns.
All that said, I wouldn't expect more than maybe a 10% performance hit
between the 2 versions (although I don't know since I've never used amber8),
as long as all other variables are kept constant. One thing I would urge
you to do is make sure that the same compiler optimization flags are present
in both versions.
If pmemd performance scales as expected (i.e. 11 is much faster than 8),
then I suppose the cost of more functionality in sander is worse
performance...
HTH,
Jason
> The results seem ok, but the calculation time does not. The problem can
> be due to a compilation problem.
>
>
>
> It seems amber 11 has been well compilated with the Intel compilator
> (gcc compilator can also be used if needed), but without using any
> special options.
>
> I would like to know if this problem is known and if there is a
> solution(compilation option, other compilator to use, ... ?).
>
>
>
> The computer I use is equipped with a Intel Xeon E5530 processor (I can
> give more details if needed)
>
>
>
> Thanks in advance,
>
> Regards,
>
> G.GUENEGOU
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon May 09 2011 - 08:30:04 PDT
