[gmx-users] How to increase the volatile gpu-util

Mark Abraham mark.j.abraham at gmail.com
Sat Jul 29 09:49:24 CEST 2017


Hi,

There is not much to be done with an RF simulation. The GPU is idle when
doing search, update and constraints. You could try a higher nstlist, as
the message reports, eg gmx mdrun -nstlist 25 but you'll only get a small
improvement.

The way to get better utilisation, if your science needs work with multiple
simulations, is to run two simulations on your node. Those will run out of
phase with eachother, slightly slower, but with nearly double the sampling
throughput. See
http://onlinelibrary.wiley.com/doi/10.1002/jcc.24030/abstract (or same on
arxiv).

Mark

On Sat, 29 Jul 2017 07:24 leila karami <karami.leila1 at gmail.com> wrote:

> Dear Mark,
>
> Thank for your answer.
>
> > Much more relevant is what gromacs reports in the end of the log file.
>
> The end of my log file is as follows:
>
> ============================================================
>
>         M E G A - F L O P S   A C C O U N T I N G
>
> NB=Group-cutoff nonbonded kernels    NxN=N-by-N cluster Verlet kernels
>  RF=Reaction-Field  VdW=Van der Waals  QSTab=quadratic-spline table
>  W3=SPC/TIP3p  W4=TIP4p (single or pairs)
>  V&F=Potential and force  V=Potential only  F=Force only
>
>  Computing:                               M-Number         M-Flops  % Flops
>
> -----------------------------------------------------------------------------
>  Pair Search distance check       388751078.685856  3498759708.173     0.3
>  NxN RF Elec. + LJ [F]            28110392002.016254 1068194896076.618
> 98.2
>  NxN RF Elec. + LJ [V&F]          283943606.686080 15332954761.048     1.4
>  Shift-X                            5089620.339308    30537722.036     0.0
>  Bonds                              2100000.007000   123900000.413     0.0
>  Angles                             1650000.005500   277200000.924     0.0
>  Impropers                           150000.000500    31200000.104     0.0
>  Virial                             5090295.339353    91625316.108     0.0
>  Stop-CM                            1017924.678616    10179246.786     0.0
>  P-Coupling                         5089620.000000    30537720.000     0.0
>  Calc-Ekin                         10179240.678616   274839498.323     0.0
>  Lincs                               525000.005250    31500000.315     0.0
>  Lincs-Mat                         10800000.108000    43200000.432     0.0
>  Constraint-V                       1050000.007000     8400000.056     0.0
>  Constraint-Vir                       26250.001750      630000.042     0.0
>
> -----------------------------------------------------------------------------
>  Total                                             1087980360051.378
>  100.0
>
> -----------------------------------------------------------------------------
>
>
>      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
>
> On 1 MPI rank, each using 16 OpenMP threads
>
>  Computing:          Num   Num      Call    Wall time         Giga-Cycles
>                      Ranks Threads  Count      (s)         total sum    %
>
> -----------------------------------------------------------------------------
>  Neighbor search        1   16   15000001  149304.198    5242873.051  11.7
>  Launch GPU ops.        1   16  300000001   17306.465     607723.034   1.4
>
>
>                                70691,2       99%
>  Neighbor search        1   16   15000001  149304.198    5242873.051  11.7
>  Launch GPU ops.        1   16  300000001   17306.465     607723.034   1.4
>  Force                  1   16  300000001  398616.578   13997571.024  31.3
>  Wait GPU local         1   16  300000001  371904.849   13059578.625  29.2
>  NB X/F buffer ops.     1   16  585000001  131860.152    4630318.829  10.4
>  Write traj.            1   16       7414    1428.772      50171.859   0.1
>  Update                 1   16  300000001  136357.663    4788250.627  10.7
>  Constraints            1   16  300000001   12172.356     427436.850   1.0
>  Rest                                       52848.935    1855810.221   4.2
>
> -----------------------------------------------------------------------------
>  Total                                     1271799.968   44659734.120 100.0
>
> -----------------------------------------------------------------------------
>
>  GPU timings
>
> -----------------------------------------------------------------------------
>  Computing:                         Count  Wall t (s)      ms/step       %
>
> -----------------------------------------------------------------------------
>  Pair list H2D                   15000001   18161.643        1.211     2.3
>  X / q H2D                      300000001  274639.405        0.915    35.2
>  Nonbonded F kernel             285000000  276274.002        0.969    35.4
>  Nonbonded F+prune k.            12000000   15443.741        1.287     2.0
>  Nonbonded F+ene+prune k.         3000001    3975.874        1.325     0.5
>  F D2H                          300000001  191562.559        0.639    24.6
>
> -----------------------------------------------------------------------------
>  Total                                     780057.223        2.600   100.0
>
> -----------------------------------------------------------------------------
>
> Force evaluation time GPU/CPU: 2.600 ms/1.329 ms = 1.957
>
> NOTE: 12 % of the run time was spent in pair search,
>       you might want to increase nstlist (this has no effect on accuracy)
>
>
>                Core t (s)   Wall t (s)        (%)
>        Time: 20365044.088  1271799.968     1601.3
>                          14d17h16:39
>                  (ns/day)    (hour/ns)
> Performance:      611.417        0.039
> Finished mdrun on rank 0 Sat Jul 29 05:40:47 2017
>
> =============================================================
>
> Best,
> --
> 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.
>


More information about the gromacs.org_gmx-users mailing list