[gmx-users] Some tips for decreasing CPU time for mdrun -rerun

[Mark Abraham]

On 20/09/2012 7:26 PM, Wu Chaofu wrote:
> Dear gmxers,
> I have generated MD simulation trajectory using gmx, and now I want to
> recalculate the energies and forces for the older trajectory by
> excluding interactions between two defined groups. Therefore, the
> older trajectory is used as one input option for mdrun through -rerun.
> In my opinions, the newer run shoud much quicker than the older one
> since only saved frames and less interactions are employed in this
> calculation. However, I find that the speed is almost same. How to
> decreasing CPU time for mdrun -rerun? I think I must have missed
> something. Could you please give me some hints? Thank you a lot for

mdrun -rerun requires neighbour searching for every frame of the 
trajectory, because it cannot assume spatial locality in the way that 
mdrun can. So it is possible for a rerun to do more work than the 
original. However in the absence of the timing breakdown reported at the 
end of the .log files, there's nothing much else to be said.

Mark