[gmx-users] Ewald sums too slow?

[Erik Lindahl]

Hi Vincent,

My guess would be you are using the default nx=ny=nz=32.

If you look at the mdp documentation it says that these numbers specify 
the number of reciprocal wave vectors in each dimension when you do 
classical Ewald, while it is the number of grid cells with PME. 
Fourierspacing doesn't enter for classical Ewald since you don't have a 
grid.

If memory serves me right (but please check the literature on this), 
reasonable "normal" numbers are in the order of 5, and in any case I'm 
pretty sure 32 is massive overkill.

We've also optimized our PME implementation quite a bit compared to Tom 
Darden's original implementation (a factor 2-3), so I would expect the 
crossover system size to be very small.

The only reason why we have classical Ewald is to check the PME 
implementation :-)

Cheers,

Erik


On Jun 15, 2004, at 7:47 PM, Vincent Ballenegger wrote:

> Hi all,
> For small systems (<100 molecules), I would expect the standard Ewald 
> sum to be faster than the PME method.
> For a system of 64 SPC water molecules in a box of width L=1.2413 nm, 
> I found however that gromacs-3.2.1 is about 85 times slower when using 
> the standard Ewald method instead of PME !!
> Why are the Ewald sums so slow?
>
> My input parameters were:
> ewald_rtol               = 1e-05
> fourierspacing = 0.12
> rlist = rcoulomb = rvdw = 0.62 nm
> pme_order                = 6
> epsilon_surface          = 0
>
> Thanks,
> Vincent
>
> _______________________________________________
> gmx-users mailing list
> gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> Please don't post (un)subscribe requests to the list. Use the www 
> interface or send it to gmx-users-request at gromacs.org.