As Daniel mentioned, it would be good if you posted some .fdf or outfiles. But if you are using only one k-point, maybe you have set ParalleloverK T, then you will not have any speed increase with more # of procs.
Regards Henrik 2010/4/15 张广平 <[email protected]> > Dear Backlund, Daniel， > I have found the reasion，it is because I set K points more than one since > the system type is a molecular.When I calculate a system with a cell big > enough that makes the system type is a molecular,I just need to set k to 1 > for each componet,i.e.Gamma point.The speed is not proportion to 1/n,with n > is the cores.when I use three or more cores,the time is about 28 seconds for > each SCF.it can not be reduced by increasing the cores. > With best wishes. > > > 在2010-04-15，"Backlund, Daniel" <[email protected]> 写道： > > > >Can you post your FDF file, an example output, and your arch.make > >file for compiling SIESTA. My first guess is that the parallel > >siesta is not being started properly. > > > >Daniel J. Backlund > >Texas Tech University - Physics > >[email protected] > > >________________________________________ > >From: [email protected] [[email protected]] On Behalf Of Guangping Zhang > >[[email protected]] > >Sent: Wednesday, April 14, 2010 10:04 PM > >To: siesta-discuss > >Subject: [SIESTA-L] about optimizing the geometry > > > >Dear siesta users and developers： > > I have found it is much slower to use siesta to optimize the geometry > > using more than one core than using only one core! > >So I wonder why? > >And another question is if I want to optimize a isolated molecular,does the > >volume of the cell affect the time used to complete the optimization,since > >the cell makes the system is molecular? > >Can DM.UseSaveDM T,MD.UseSaveZM T,MD.UseSaveCG T make a whole > >continue start from previous run,if I use %block Zmatrix as the geometry > >input in the input.fdf? > >Is that enought? > >Guangping Zhang > > > >
