[gmx-users] Slow in double-precision version

[Mark Abraham]

On 25/10/2011 3:11 PM, Sai Pooja wrote:
> Thank you so much Mark. You are right about the cut-off values. I am
> now using 1.2nm which appears more commonly in literature. If possible
> please take a look at these statistics:
>
>       R E A L   C Y C L E   A N D   T I M E   A C C O U N T I N G
>
>   Computing:         Nodes     Number     G-Cycles    Seconds     %
> -----------------------------------------------------------------------
>   Neighbor search        1     380527   166719.792    59692.7     2.3
>   Force                  1    3805262  1068307.182   382498.9    14.8
>   Write traj.            1      10433     1319.463      472.4     0.0
>   Update                 1    3805262    12043.902     4312.2     0.2
>   Constraints            1    3805262    18769.536     6720.3     0.3
>   Rest                   1             5965690.942  2135968.4    82.5
> -----------------------------------------------------------------------
>   Total                  1             7232850.817  2589664.9   100.0
> -----------------------------------------------------------------------
>
>                 NODE (s)   Real (s)      (%)
>         Time: 2476509.010 2589664.940     95.6
>                         28d15h55:09
>                 (Mnbf/s)   (MFlops)   (ns/day)  (hour/ns)
> Performance:     16.281    468.817      0.266     90.391
> Finished mdrun on node 0 Thu Oct 13 21:43:03 2011
>
> It seems like "Rest" option is taking 82% time. Now would this
> correspond to an external program like plumed

Yep. "rest" is a catch-all, and plumed is probably not anywhere else. 
Their developers would be able to tell you.

Mark

> or is it something else?
>
> Regards
> Pooja
>
>
> On Mon, Oct 24, 2011 at 8:18 PM, Mark Abraham<Mark.Abraham at anu.edu.au>  wrote:
>> On 25/10/2011 9:54 AM, Sai Pooja wrote:
>>> Hi all,
>>>
>>> I am using GROMACS with the plugin PLUMED in double precision. The
>>> speed has gone down by a huge margin and I was wondering if there is
>>> something in 1) compilation 2) md parameters that could be
>>> contributing to this?
>> Simply moving to double precision will slow things down noticeably. How much
>> depends on a lot of things, but you can test this yourself with a non-PLUMED
>> double-precision GROMACS.
>>
>> Otherwise, you should consider whether whatever you are doing with PLUMED is
>> expected to take a lot longer than normal...
>>
>>> md parameters:
>>>
>>>
>>> title                    = NVT simulation (constant number, volume and
>>> temperature)
>>> cpp                      = /lib/cpp
>>> define                   = -DPOSRES
>>>
>>> ; RUN CONTROL PARAMETERS
>>> integrator               = md
>>> dt                       = 0.002
>>> nsteps                   = 5000000
>>>
>>> ; OUTPUT CONTROL OPTIONS
>>> nstxout                  = 0 ; ast frame (coordinates)
>>> nstvout                  = 2000   ;  (velocities)
>>> nstfout                  = 2000   ; No output, except for last frame
>>> (forces)
>>> nstlog                   = 1000  ; Write every step to the log
>>> nstenergy                = 1000  ; Write energies at every step
>>> xtc_grps                 = Protein Non-Protein
>>> nstxtcout                = 500   ; Do not write a compressed trajectory
>>> energygrps               = Protein SOL  ; Write energy information
>>> separately for these groups
>>> ;energygrp_table          = Protein SOL SOL SOL
>>>
>>> ; NEIGHBORSEARCHING PARAMETERS
>>> nstlist                  = 10
>>> ns-type                  = Grid
>>> pbc                      = xyz
>>> rlist                    = 1.8
>>>
>>> ; OPTIONS FOR ELECTROSTATICS AND VDW
>>> coulombtype              = Cut-off
>>> fourierspacing           = 0.15
>>> rcoulomb                 = 1.8
>>> ;rcoulomb_switch         = 1.6
>>> epsilon_rf               = 78
>>> vdw-type                 = Cut-off
>>> rvdw                     = 1.8
>>> ;rvdw-switch             = 1.6
>>> ;table-extension                 = 1.0
>>>
>>> ; FFT grid size, when a value is 0 fourierspacing will be used =
>>> fourier_nx               = 0
>>> fourier_ny               = 0
>>> fourier_nz               = 0
>>> ; EWALD/PME/PPPM parameters =
>>> pme_order                = 4
>>> ewald_rtol               = 1e-05
>>> epsilon_surface          = 0
>>> optimize_fft             = no
>>> ; Temperature coupling
>>> tcoupl                   = nose-hoover
>>> tc-grps                  = Protein  Non-Protein
>>> tau_t                    = 0.4      0.4
>>> ref_t                    = 300 300
>>>
>>> ; Pressure coupling
>>> pcoupl                   = no
>>>
>>> ; OPTIONS FOR BONDS
>>> constraints              = all-bonds
>>>
>>>
>>> Details of the simulation:
>>> 1) MD with explicit solvent using charmm27 with tip3p (recommended one)
>> Those cut-offs do not look like any I've ever seen for CHARMM force fields.
>> It is not wise to haphazardly choose large values for poor electrostatic
>> algorithms and hope for the best - even though almost all force fields were
>> parameterized with poor electrostatic treatments.
>>
>> Mark
>>
>>> 2) gromacs 4.5.1
>>> 3) With plumed
>>> 4) Double precision
>>> 5) Position restraints for some parts of the protein (5 chains)
>>>
>>>
>>> I would really appreciate any help.
>>>
>>>
>>> Regards
>>> Pooja
>>>
>>>
>>>
>>> Pooja
>>>
>> --
>> gmx-users mailing list    gmx-users at gromacs.org
>> http://lists.gromacs.org/mailman/listinfo/gmx-users
>> Please search the archive at
>> http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
>> Please don't post (un)subscribe requests to the list. Use the www interface
>> or send it to gmx-users-request at gromacs.org.
>> Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>
>
>