[gmx-users] difference of calculation time between the numbers of fep-lambdas values

nao.morishita at takeda.com nao.morishita at takeda.com
Tue Apr 28 02:34:55 CEST 2015


Hi Justin and Hannes,



Thank you for the replay.

I apologize for the insufficient information.

Actually, I calculated them with couple-moltype = solute and init-lambda-state = 1.

The other parameters except for fep-lambdas, the number of core(8) and hardware were same.

When I look at "REAL CCLE AND TIME ACCOUNTING" in log file, there is large difference in "PME redist. X/F".



What makes it much longer?

I would appreciate any comments or concerns.



Input1(fep-lambdas           = 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0)

****************************************************************************************



Computing:         Nodes   Th.     Count  Wall t (s)     G-Cycles       %

-----------------------------------------------------------------------------

Domain decomp.         8    1     250000     100.985     2100.559     0.2

DD comm. load          8    1        501       0.004        0.082     0.0

Neighbor search        8    1     250001     351.227     7305.783     0.8

Comm. coord.           8    1    2500001      74.248     1544.405     0.2

Force                  8    1    2500001    8671.720   180378.288    20.8

Wait + Comm. F         8    1    2500001      49.926     1038.492     0.1

PME mesh               8    1    2500001   31496.259   655145.855    75.6

Write traj.            8    1        547       0.275        5.710     0.0

Update                 8    1    2500001     284.487     5917.540     0.7

Constraints            8    1    5000002     396.070     8238.552     1.0

Comm. energies         8    1     250001       7.876      163.837     0.0

Rest                   8                     250.084     5201.928     0.6

-----------------------------------------------------------------------------

Total                  8                   41683.159   867041.031   100.0

-----------------------------------------------------------------------------

-----------------------------------------------------------------------------

PME redist. X/F        8    1    7500003   26513.015   551490.632    63.6

PME spread/gather      8    1   10000004    3879.936    80705.579     9.3

PME 3D-FFT             8    1   10000004     702.656    14615.765     1.7

PME 3D-FFT Comm.       8    1   20000008     264.536     5502.543     0.6

PME solve              8    1    5000002     127.236     2646.610     0.3

-----------------------------------------------------------------------------



               Core t (s)   Wall t (s)        (%)

       Time:   332678.400    41683.159      798.1

                         11h34:43

                 (ns/day)    (hour/ns)

Performance:       10.364        2.316

Finished mdrun on node 0 Thu Apr  2 06:31:45 2015

*********************************************************************************

Input2(fep-lambdas           = 0.0 0.1 0.2)

*********************************************************************************

Computing:         Nodes   Th.     Count  Wall t (s)     G-Cycles       %

-----------------------------------------------------------------------------

Domain decomp.         8    1     250000      85.779     1784.259     0.5

DD comm. load          8    1        501       0.003        0.072     0.0

Neighbor search        8    1     250001     312.515     6500.517     1.7

Comm. coord.           8    1    2500001      69.584     1447.387     0.4

Force                  8    1    2500001    4219.963    87778.089    23.5

Wait + Comm. F         8    1    2500001      46.401      965.178     0.3

PME mesh               8    1    2500001   12342.326   256728.717    68.8

Write traj.            8    1        520       0.150        3.117     0.0

Update                 8    1    2500001     262.237     5454.706     1.5

Constraints            8    1    5000002     368.111     7656.961     2.1

Comm. energies         8    1     250001       6.524      135.712     0.0

Rest                   8                     227.039     4722.575     1.3

-----------------------------------------------------------------------------

Total                  8                   17940.633   373177.290   100.0

-----------------------------------------------------------------------------

-----------------------------------------------------------------------------

PME redist. X/F        8    1    7500003    7982.056   166032.158    44.5

PME spread/gather      8    1   10000004    3327.975    69224.129    18.5

PME 3D-FFT             8    1   10000004     645.036    13417.186     3.6

PME 3D-FFT Comm.       8    1   20000008     259.876     5405.587     1.4

PME solve              8    1    5000002     118.906     2473.318     0.7

-----------------------------------------------------------------------------



               Core t (s)   Wall t (s)        (%)

       Time:   143196.260    17940.633      798.2

                         4h59:00

                 (ns/day)    (hour/ns)

Performance:       24.079        0.997

Finished mdrun on node 0 Fri Apr  3 19:11:15 2015

*********************************************************************************



Best regards,



Nao





-----Original Message-----
From: gromacs.org_gmx-users-bounces at maillist.sys.kth.se [mailto:gromacs.org_gmx-users-bounces at maillist.sys.kth.se] On Behalf Of Hannes Loeffler
Sent: Monday, April 27, 2015 10:48 PM
To: gromacs.org_gmx-users at maillist.sys.kth.se
Cc: gmx-users at gromacs.org
Subject: Re: [gmx-users] difference of calculation time between the numbers of fep-lambdas values



On Mon, 27 Apr 2015 09:05:05 -0400

Justin Lemkul <jalemkul at vt.edu<mailto:jalemkul at vt.edu>> wrote:



>

>

> On 4/27/15 9:02 AM, Hannes Loeffler wrote:

> > On Mon, 27 Apr 2015 07:43:51 -0400

> > Justin Lemkul <jalemkul at vt.edu<mailto:jalemkul at vt.edu>> wrote:

> >

> >> Note that without couple-moltype, you're going to be decoupling the

> >> whole system, which is (1) not what you want for calculating

> >> solvation free energy and (2) extremely slow.  There is an inherent

> >> slowdown when running the free energy code, but it should not be so

> >> large.

> >

> > Really? I just looked into the code yesterday and I think that if

> > you do not set couple-moltype, none of the coupling code is ever

> > triggered.  I have only played a bit with relative free energies

> > (Gromacs 4.6) where I think you don't need any of the couple-

> > parameters at all.

> >

>

> Maybe I got that backwards.  For relative free energies, this is true

> (we're doing these now); topological differences are all that are

> needed.  For an absolute solvation free energy, one needs

> couple-moltype unless the B-state (dummy) is explicitly defined in

> the topology, AFAIK.  If there is no couple-moltype, and no B-state

> defined in the topology, I'm not sure what the code would even be

> doing, actually.



I think it runs a normal MD simulation for each lambda.  A quick test

with a A-state only topology and no couple- parameters gives only

gradients with zero value.  In fact, the output file says



There are 0 atoms and 0 charges for free energy perturbation





Cheers,

Hannes.



--

Gromacs Users mailing list



* Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting!



* Can't post? Read http://www.gromacs.org/Support/Mailing_Lists



* For (un)subscribe requests visit

https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-request at gromacs.org<mailto:gmx-users-request at gromacs.org>.


More information about the gromacs.org_gmx-users mailing list