[gmx-users] freezing atom group speed up issue
David
spoel at xray.bmc.uu.se
Wed Jul 14 22:29:04 CEST 2004
On Wed, 2004-07-14 at 21:59, Sung-Joo Lee wrote:
> Hi,
>
> I've been observing the speed up gained by freezing atoms for my 120k
> atom system (20k solvent, 100k protein cage surrounding solvent)
>
> Freezing the protein cage (100k out of 120k) had things moving twice as
> fast.. slower than I was hoping it to be.
>
> Basically the freeze group setting killed most of the LJ(T) + Coul(T),
> (T=Tabulated), and half of Innerloop-Iatom, and some of Coulomb(T) but
> the rest (NS-pairs, RB-Dihedrals, Angles) remain the same.
>
> .mdp setting related to the freezing was. (Dummy is a single atom group,
> and Cage consists ~100k atoms)
>
> freezegrps = Dummy Cage
> freezedim = Y Y Y Y Y Y
> energygrp_excl = Cage Cage
> energygrps = Cage
>
> Am I missing something here?
>
> Perhaps I could comb through the topology to remove the bonded
> interactions within the freeze group
>
> Any suggestions on how to get it even faster?
>
As you say, filtering out interactions with frozen atoms. It would
require something more than just the numbers presented by mdrun. You'd
have to compare some profiling data.
Neighborsearching is (always) a good candidate for optimization. On the
other hand, while NS maybe slow, the reason for this is that it takes
the complexity out of the innerloops which are fast.
Bonded interactions (and constraints) with only frozen atoms can be
filtered out, preferably by grompp. If you feel like coding that, please
go ahead...
> Thanks,
>
> Sungjoo Lee
>
> Postdoc, Stanford University.
> http://folding.stanford.edu
>
>
> Details follows
>
> None of 120k atoms freezed
>
> RF=Reaction-field Free=Free Energy SC=Softcore
> T=Tabulated S=Solvent W=Water WW=Water-Water
>
> Computing: M-Number M-Flop's % Flop's
> LJ + Coulomb 261.486225 9936.476550 0.7
> Coulomb(T) 1100.638222 45126.167102 3.0
> Coulomb(T)(W) 7.352676 904.379148 0.1
> LJ(T) + Coul(T) 16085.385060 1077720.799020 71.3
> LJ(T) + Coul(T)(W) 435.516667 64891.983383 4.3
> LJ(T) + Coul(T)(WW) 319.692207 123401.191902 8.2
> Innerloop-Iatom 259.895420 2598.954200 0.2
> NS-Pairs 4145.122210 87047.566410 5.8
> Reset In Box 12.069096 108.621864 0.0
> Shift-X 239.228990 1435.373940 0.1
> CG-CoM 10.711656 310.638024 0.0
> Bonds 50.620570 2176.684510 0.1
> Angles 181.145965 29526.792295 2.0
> Propers 19.756737 4524.292773 0.3
> RB-Dihedrals 169.872703 41958.557641 2.8
> Virial 119.642523 2153.565414 0.1
> Update 119.615496 3708.080376 0.2
> Stop-CM 119.496000 1194.960000 0.1
> Calc-Ekin 119.734992 3232.844784 0.2
> Lincs 49.732752 2983.965120 0.2
> Lincs-Mat 265.790988 1063.163952 0.1
> Shake-V 119.615496 1794.232440 0.1
> Shake-Vir 119.615496 2153.078928 0.1
> Settle 6.740160 2177.071680 0.1
> Total 1512129.44146 100.0
>
> NODE (s) Real (s) (%)
> Time: 1635.680 1636.000 100.0
> 27:15
> (Mnbf/s) (MFlops) (ps/NODE hour) (NODE hour/ns)
> Performance: 13.238 924.465 4.402 227.178
>
>
>
>
> 100k of 120k atoms freezed (most protein freezed, and solvent free)
>
> Computing: M-Number M-Flop's % Flop's
> LJ + Coulomb 261.486225 9936.476550 2.1
> Coulomb(T) 948.149043 38874.110763 8.2
> Coulomb(T)(W) 7.476621 919.624383 0.2
> LJ(T) + Coul(T) 576.356673 38615.897091 8.1
> LJ(T) + Coul(T)(W) 447.212200 66634.617800 14.0
> LJ(T) + Coul(T)(WW) 339.711626 131128.687636 27.6
> Innerloop-Iatom 136.939390 1369.393900 0.3
> NS-Pairs 4166.729882 87501.327522 18.4
> Reset In Box 12.069096 108.621864 0.0
> Shift-X 239.228990 1435.373940 0.3
> CG-CoM 10.711656 310.638024 0.1
> Bonds 50.620570 2176.684510 0.5
> Angles 181.145965 29526.792295 6.2
> Propers 19.756737 4524.292773 1.0
> RB-Dihedrals 169.872703 41958.557641 8.8
> Virial 119.642523 2153.565414 0.5
> Update 119.615496 3708.080376 0.8
> Stop-CM 119.496000 1194.960000 0.3
> Calc-Ekin 119.734992 3232.844784 0.7
> Lincs 49.732752 2983.965120 0.6
> Lincs-Mat 265.790988 1063.163952 0.2
> Shake-V 119.615496 1794.232440 0.4
> Shake-Vir 119.615496 2153.078928 0.5
> Settle 6.740160 2177.071680 0.5
> Total 475482.05939 100.0
>
> NODE (s) Real (s) (%)
> Time: 817.810 818.000 100.0
> 13:37
> (Mnbf/s) (MFlops) (ps/NODE hour) (NODE hour/ns)
> Performance: 7.590 581.409 8.804 113.585
> _______________________________________________
> 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.
--
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
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
More information about the gromacs.org_gmx-users
mailing list