[gmx-users] hardware problem of GPU?

Mark Abraham mark.j.abraham at gmail.com
Fri Jun 6 14:08:50 CEST 2014


There's only one rank, so you can only be using one GPU!

Mark
On Jun 6, 2014 8:13 AM, "Albert" <mailmd2011 at gmail.com> wrote:

> Hi Mark:
>
> thanks a lot for reply.  Here is my log file informations.  I've got
> another GPU machine with two GTX690, and the double CPU job is much faster
> than single GPU. But this dual GTX780Ti is not the case, so I carious about
> what's happening to the hardware since Gromacs was compiled in the same
> way, and the testing system are the same.
>
> thanks a lot
>
> -----------------------------------------------log----------
> ------------------------------------------------------------
> -----------------------
>  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          449758.183440 4047823.651     0.1
>  NxN Ewald Elec. + VdW [F]        114203606.933184 7537438057.590    95.3
>  NxN Ewald Elec. + VdW [V&F]        1153624.365888 123437807.150     1.6
>  1,4 nonbonded interactions           30707.512283 2763676.105     0.0
>  Calc Weights                        413752.665501 14895095.958     0.2
>  Spread Q Bspline                   8826723.530688 17653447.061     0.2
>  Gather F Bspline                   8826723.530688 52960341.184     0.7
>  3D-FFT                            15297568.453746 122380547.630     1.5
>  Solve PME                             7839.867456 501751.517     0.0
>  Shift-X                               3447.992667 20687.956     0.0
>  Angles                               21342.508537 3585541.434     0.0
>  Propers                              32957.513183 7547270.519     0.1
>  Impropers                             3147.501259 654680.262     0.0
>  RB-Dihedrals                            87.500035 21612.509     0.0
>  Virial                               13803.055212 248454.994     0.0
>  Stop-CM                               1379.285334 13792.853     0.0
>  Calc-Ekin                            27583.610334 744757.479     0.0
>  Lincs                                11865.004746 711900.285     0.0
>  Lincs-Mat                           256110.102444 1024440.410     0.0
>  Constraint-V                        149670.059868 1197360.479     0.0
>  Constraint-Vir                       13780.555122 330733.323     0.0
>  Settle                               41980.016792 13559545.424     0.2
> ------------------------------------------------------------
> -----------------
>  Total                                              7905739325.773 100.0
> ------------------------------------------------------------
> -----------------
>
>
>      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
>
>  Computing:         Nodes   Th.     Count  Wall t (s) G-Cycles       %
> ------------------------------------------------------------
> -----------------
>  Neighbor search        1   20      62501     156.247 9380.162     2.3
>  Launch GPU ops.        1   20    2500001     182.404 10950.541     2.7
>  Force                  1   20    2500001    1047.581 62890.858    15.3
>  PME mesh               1   20    2500001    2546.280 152864.323    37.3
>  Wait GPU local         1   20    2500001     808.773 48554.193    11.8
>  NB X/F buffer ops.     1   20    4937501     114.557 6877.380     1.7
>  Write traj.            1   20         58       1.380 82.874     0.0
>  Update                 1   20    2500001     519.331 31177.740     7.6
>  Constraints            1   20    2500001     757.477 45474.638    11.1
>  Rest                   1                     694.482 41692.777    10.2
> ------------------------------------------------------------
> -----------------
>  Total                  1                    6828.512   409945.484 100.0
> ------------------------------------------------------------
> -----------------
> ------------------------------------------------------------
> -----------------
>  PME spread/gather      1   20    5000002    1910.053 114668.815    28.0
>  PME 3D-FFT             1   20    5000002     516.241 30992.236     7.6
>  PME solve              1   20    2500001     112.115 6730.761     1.6
> ------------------------------------------------------------
> -----------------
>
>  GPU timings
> ------------------------------------------------------------
> -----------------
>  Computing:                         Count  Wall t (s) ms/step       %
> ------------------------------------------------------------
> -----------------
>  Pair list H2D                      62501      14.934 0.239     0.3
>  X / q H2D                        2500001     206.939 0.083     4.6
>  Nonbonded F kernel               2250000    3527.275 1.568    78.8
>  Nonbonded F+ene k.                187500     405.370 2.162     9.1
>  Nonbonded F+ene+prune k.           62501     167.980 2.688     3.8
>  F D2H                            2500001     154.048 0.062     3.4
> ------------------------------------------------------------
> -----------------
>  Total                                       4476.545        1.791 100.0
> ------------------------------------------------------------
> -----------------
>
> Force evaluation time GPU/CPU: 1.791 ms/1.438 ms = 1.246
> For optimal performance this ratio should be close to 1!
>
>
> NOTE: The GPU has >20% more load than the CPU. This imbalance causes
>       performance loss, consider using a shorter cut-off and a finer PME
> grid.
>
>                Core t (s)   Wall t (s)        (%)
>        Time:   136384.758     6828.512     1997.3
>                          1h53:48
>                  (ns/day)    (hour/ns)
> Performance:       63.264        0.379
>
>
>
>
> On 06/05/2014 10:05 PM, Mark Abraham wrote:
>
>> What did you learn from the performance output at the end of the log file?
>>
>> Mark
>>
>
> --
> 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