[gmx-users] Should I use separate PME nodes

Carsten Kutzner ckutzne at gwdg.de
Fri Jun 25 15:42:59 CEST 2010


Hi Gaurav,

separate PME nodes usually pay off on a larger number of nodes
(>16). In rare cases, you will see a performance benefit on a small number
of nodes as well. Just try it! Or use g_tune_pme ... ;)

Carsten


On Jun 25, 2010, at 3:32 PM, Gaurav Goel wrote:

> I ran my simulation in parallel on 4 nodes (with zero separate PME
> nodes). Below is the information printed in md.log.
> 
> I see that PME-Mesh calculations took 60% of CPU time. Any
> recommendations on using 1 or more separate PME nodes to speed up?
> 
> 
> Computing:                         M-Number         M-Flops  % Flops
> -----------------------------------------------------------------------
> Coul(T) + VdW(T)             1761401.496982   119775301.795    20.2
> Outer nonbonded loop          106414.135764     1064141.358     0.2
> Calc Weights                   32400.006480     1166400.233     0.2
> Spread Q Bspline             2332800.466560     4665600.933     0.8
> Gather F Bspline             2332800.466560    27993605.599     4.7
> 3D-FFT                      47185929.437184   377487435.497    63.6
> Solve PME                     675840.135168    43253768.651     7.3
> NS-Pairs                      823453.927656    17292532.481     2.9
> Reset In Box                    2160.002160        6480.006     0.0
> CG-CoM                          2160.004320        6480.013     0.0
> Virial                         11700.002340      210600.042     0.0
> Ext.ens. Update                10800.002160      583200.117     0.1
> Stop-CM                        10800.002160      108000.022     0.0
> Calc-Ekin                      10800.004320      291600.117     0.0
> -----------------------------------------------------------------------
> Total                                         593905146.863   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     Number     G-Cycles    Seconds     %
> -----------------------------------------------------------------------
> Domain decomp.         4    1000001     3859.416     1488.1     0.6
> Comm. coord.           4    5000001     1874.635      722.8     0.3
> Neighbor search        4    1000001    78640.722    30322.2    11.2
> Force                  4    5000001   180659.902    69658.5    25.8
> Wait + Comm. F         4    5000001     2578.994      994.4     0.4
> PME mesh               4    5000001   422268.834   162817.7    60.4
> Write traj.            4      10001       17.526        6.8     0.0
> Update                 4    5000001     2981.794     1149.7     0.4
> Comm. energies         4    5000001     2633.176     1015.3     0.4
> Rest                   4                3580.341     1380.5     0.5
> -----------------------------------------------------------------------
> Total                  4              699095.342   269556.0   100.0
> -----------------------------------------------------------------------
> 
> Thanks,
> Gaurav
> -- 
> 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/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/mailing_lists/users.php


--
Dr. Carsten Kutzner
Max Planck Institute for Biophysical Chemistry
Theoretical and Computational Biophysics
Am Fassberg 11, 37077 Goettingen, Germany
Tel. +49-551-2012313, Fax: +49-551-2012302
http://www.mpibpc.mpg.de/home/grubmueller/ihp/ckutzne







More information about the gromacs.org_gmx-users mailing list