[Wien] lapw0_mpi for dual cores

[Gerhard Fecher]

I forgot to mention OMP_NUM_THREADS=2 works only for paralellised parts of the
program, that is usually those that make excessive use of the mkl that has already
parallel routines.

Thanks for that test, I wonder that the speed increases that few, I have other band structure programs
that gain up to 30% speed. I did not use -parallel up to now, because I do oftenly quasi-parallel runs by hand. For example
if optimizing a small cell, I do one run for even (-4%,-2,0, etc) changes and another one at the same time for odd changes
of the lattice parameter (-3%,-1, etc). Seems, this makes optimal use of the deadtime of the second CPU. But indeed, it may be a bad idea if using large systems.

I think Peter will know best how and where to gain speed.

Ciao
Gerhard


-----Ursprüngliche Nachricht-----
Von: wien-bounces at zeus.theochem.tuwien.ac.at im Auftrag von Michael Gurnett
Gesendet: Sa 09.09.2006 09:05
An: A Mailing list for WIEN2k users
Betreff: Re: [Wien] lapw0_mpi for dual cores
 
Tried that now, and the reulst is that with -parallel there is a 20
second speed increase for a 44 min lapw0 run. So I would say the
increase is quite insignificant. Still it was worth a try.

Michael


On Fri, 2006-09-08 at 19:23 +0200, Gerhard H Fecher wrote:
> Did you try to compile simply with the auto paralelliser:  -parallel 
> 
> I did not check it up to know but would be interesting if it speeds up on dual 
> core machines without using special mpi libraries, what would be easier for 
> unexperienced users.
> 
> Ciao
> Gerhard
> 
> Am Freitag, 8. September 2006 18:29 schrieb Michael Gurnett:
> > I've been checking the system monitor and it appears that even with
> > OMP_NUM_THREADS=2 lapw0 really only using about 50% of the processing
> > power of the dual cores. Has anyone tried using lapw0_mpi for a dual
> > core, and was it easy to implement
> > 
> > 
> > Thanks 
> > 
> > Michael
> > 
> > _______________________________________________
> > Wien mailing list
> > Wien at zeus.theochem.tuwien.ac.at
> > http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> > 
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> 

_______________________________________________
Wien mailing list
Wien at zeus.theochem.tuwien.ac.at
http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/ms-tnef
Size: 3888 bytes
Desc: not available
Url : http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20060909/f8b0140d/attachment.bin