[gmx-users] Excessive compute time using PME

[David van der Spoel]

On Tue, 2005-05-17 at 00:52 +0200, Jordi Camps wrote:
> Hi all!
> 
> I've tried turning off the optimize_fft parameter, and the running time
> changed to 140s, while when running with optimize_fft=yes the time was 141s.
> It doesn't make a big change.
> 
> The grid generated is:
> 	Using a fourier grid of 567x539x462, spacing 0.119 0.118 0.119
> which has the next dimensions:
> 	567==(3^4)*7
> 	539==(7^2)*11
> 	462==2*3*7*11

This means your box is 60 nm. Is that correct?

> All of them can be computed with the default FFT optimizations of the FFTW
> libraries. How the grid should look like?
> 
> Anyone has another suggestion? This long times are killing us :-(
> 
> Thanks for all,
> 
> -- 
> Jordi Camps Puchades
> Instituto Nacional de Bioinformatica (INB) Nodo Computacional GNHC-2
> UPC-CIRI
> c/. Jordi Girona 1-3              
> Modul C6-E201                     Tel.  : 934 011 650
> E-08034 Barcelona                 Fax   : 934 017 014
> Catalunya (Spain)                 e-mail: jcamps at lsi.upc.edu
> 
> -----Mensaje original-----
> De: gmx-users-bounces at gromacs.org [mailto:gmx-users-bounces at gromacs.org] En
> nombre de David
> Enviado el: sábado, 14 de mayo de 2005 9:24
> Para: Discussion list for GROMACS users
> Asunto: RE: [gmx-users] Excessive compute time using PME
> 
> 
> On Sat, 2005-05-14 at 03:29 +0200, Jordi Camps wrote:
> > Hello!
> > 
> > The box is full of waters (plus the protein) and not specially large. 
> > It is a truncated octahedron box. The grid spacing is the standard:
> > 	fourierspacing=0.12
> 
> Check the grid anyway, but the first steps are extra costly because you
> turned on optimize_fft. Try it without.
> 
> > Either way, this line is commented out.
> > 
> > The parameters are:
> > 	title               =  Simulation
> > 	cpp                 =  /lib/cpp
> > 	integrator          =  md
> > 	dt                  =  0.002
> > 	nsteps              =  2
> > 	nstlog              =  500
> > 	nstxout             =  500
> > 	nstenergy           =  500
> > 	nstvout             =  500
> > 	nstxtcout           =  500
> > 	constraint_algorithm=  shake
> > 	constraints         =  hbonds
> > 	pbc                 =  xyz
> > 
> > 	;;;; Pressure coupling is on
> > 	Pcoupl              =  berendsen
> > 	pcoupltype          =  isotropic
> > 	;;;;tau_p               =  0.5
> > 	compressibility     =  4.5e-5
> > 	ref_p               =  1.01325
> > 
> > 	;;;; Berendsen temperature coupling is on in three groups
> > 	Tcoupl              =  berendsen
> > 	tau_t               =  0.1
> > 	tc_grps             =  System
> > 	ref_t               =  300
> > 
> > 
> > 	coulombtype         =  PME
> > 	rcoulomb            =  1.0
> > 	rvdw                =  1.0
> > 	rlist               =  1.0
> > 	optimize_fft        =  yes
> > 	;;;; Generate velocites is on at 300 K.
> > 	gen_vel             =  no
> > 
> > Thanks for your help,
> > 
> > --
> > Jordi Camps Puchades
> > Instituto Nacional de Bioinformatica (INB) Nodo Computacional GNHC-2
> > UPC-CIRI
> > c/. Jordi Girona 1-3              
> > Modul C6-E201                     Tel.  : 934 011 650
> > E-08034 Barcelona                 Fax   : 934 017 014
> > Catalunya (Spain)                 e-mail: jcamps at lsi.upc.edu
> > 
> > -----Mensaje original-----
> > De: gmx-users-bounces at gromacs.org 
> > [mailto:gmx-users-bounces at gromacs.org] En nombre de David Enviado el: 
> > viernes, 13 de mayo de 2005 19:20
> > Para: Discussion list for GROMACS users
> > Asunto: Re: [gmx-users] Excessive compute time using PME
> > 
> > 
> > On Fri, 2005-05-13 at 18:05 +0200, Jordi Camps wrote:
> > > Dear sirs,
> > > 
> > > I'm trying to run some simulations with PME, but the time that them
> > > takes are unusually long. Finally I reduced the number of steps to 
> > > compute to only two (2) steps and one system with about 20000 atoms 
> > > takes about 140 seconds to compute (spending 98% of the time "solving 
> > > PME"). But if I take out the coulombtype=PME parameter (so I simulate 
> > > with a cut-off), the time needed for the same two steps is about 3 
> > > seconds. Adding PME calculus should increase the compute time, but not 
> > > in this magnitude order.
> > > If I let the simulation run for more time, I get about 1500 steps in 12
> > > hours (approx.).
> > > 
> > > Is this normal? Could it be a compilation problem? An uncorrected 
> > > bug?
> > > A parameter definition problem?
> > 
> > Is your box nearly empty and very large? Or are you using a very fine 
> > spaced grid?
> > 
> > > 
> > > I can supply the .mdp and the other files if necessary.
> > > 
> > > I've been simulating on one processor (in multiple processors 
> > > behaves
> > > similarly). I'm using PBC too.
> > > 
> > > Thanks for your help.
> > > 
> > > Sincerely,
> > > 
-- 
David.
________________________________________________________________________
David van der Spoel, PhD, Assoc. Prof., Molecular Biophysics group,
Dept. of Cell and Molecular Biology, Uppsala University.
Husargatan 3, Box 596,          75124 Uppsala, Sweden
phone:  46 18 471 4205          fax: 46 18 511 755
spoel at xray.bmc.uu.se    spoel at gromacs.org   http://xray.bmc.uu.se/~spoel
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++