Convergence problem on CRAY-XT4

[Jun]

It doesn't work for 4, 8 and 16 processors.
Jun

On Aug 1, 3:20 pm, "Shawn T. Brown" <shawn... at gmail.com> wrote:
> Just out of curiosity, does it work on 8?  Could there be a memory overrun
> that is alleviated by making a distributed array smaller?
>
> On Fri, Aug 1, 2008 at 10:19 AM, Jun <chen... at googlemail.com> wrote:
>
> > Hi Matt,
>
> > You are right. It works when using 32 processors.
>
> > Jun
>
> > On Aug 1, 3:08 pm, Matt W <MattWa... at gmail.com> wrote:
> > > It does work for me if you run on 32 rather than 16 processors!  I'll
> > > keep looking, but I'm afraid no definite answer yet.
>
> > > Matt
>
> > > On Aug 1, 2:54 pm, Jun <chen... at googlemail.com> wrote:
>
> > > > Hi Matt,
>
> > > > Thank you for the help.
> > > > I tried your suggestion. Unfortunately, it doesn't work. The output
> > > > files are
> > > > almost the same before and after adding the section.
> > > > Any other suggestions? Thanks.
>
> > > > Jun
>
> > > > On Aug 1, 2:28 pm, Matt W <MattWa... at gmail.com> wrote:
>
> > > > > Hi Jun,
>
> > > > > there's definitely something wrong here - I can reproduce your
> > problem
> > > > > with my executable.
>
> > > > > For the time being try adding
>
> > > > >       &RS_GRID
> > > > >        DISTRIBUTION_TYPE REPLICATED
> > > > >       &END RS_GRID
>
> > > > > into the &MGRID section and see if you get stable MD.
>
> > > > > Matt
>
> > > > > On Aug 1, 12:25 pm, Jun <chen... at googlemail.com> wrote:
>
> > > > > > Hi Matt,
>
> > > > > > I just tried to run your executable, but I cannot access it.
> > > > > > I will send you my input file to your email. Could you run it
> > please.
> > > > > > Thanks a lot.
>
> > > > > > Jun
>
> > > > > > On Aug 1, 11:24 am, Matt W <MattWa... at gmail.com> wrote:
>
> > > > > > > There are four people I can see currently running cp2k on Hector
> > ...
> > > > > > > so hopefully we're not all wrong.
>
> > > > > > > You can try my executable /home/e05/e05/uclmbw/temp/cp2k.popt
>
> > > > > > > and see if that also does the same (or put files for me to try
> > your
> > > > > > > run in that directory).
>
> > > > > > > Matt