how to improve CP2K scaling for small systems

[Matt W]

Hi again,

those data seem show reasonable scaling across the cores in the box - 
certainly a big improvement on 2 cores compared to 1 core ...

1 core  CP2K                                 1  1.0    0.483    0.483 
1812.510 1812.510
2 core  CP2K                                 1  1.0    0.322    0.329 
1075.661 1075.664
4 core  CP2K                                 1  1.0    0.182    0.183 
 753.037  753.042

which sort of suggests that the communications between nodes is getting 
saturated, rather than the inbox being the problem?

It might be clearer still with the same data for the 4 vs 8 vs16 MPI 
processors on the cluster.

Matt


On Monday, March 17, 2014 11:12:45 AM UTC, labicia wrote:
>
> Dear all,
> thank you very much for the help.
>
> Please find attached a file with the timings for the same run done with:
> 1 MPI
> 2 MPI
> 4 MPI
> on a single i5 2550K processor. The machine has 16GB of ram and the memory 
> required by the simulation is about 12GB, no swap.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20140317/012adb3c/attachment.htm>