[Wien] VLM's (very large matrices)

[Peter Blaha]

Well, PCs come to some limit and sequential codes too.

Such an example (if it really requires such large matrices ?) would be worth
to try with the mpi-parallel version on a shared memory maschine.
(I recommend to use at least 4 processors, because the mpi version will
require twice as much memory)

Of course I undestand that installing scalapack-pblas is not so simple,
but that's the way to go.

> Dear WIEN users,
> I'm using the wien code compiled with ifc7 and statically linked with mkl6, and I'm trying to push nmatmax as far as it will go.  I get crashes (for real system, ie lapw1) above about nmatmax = 12500 (nume is set to 2000 - okay, for systems with few electrons I could gain a little there), corresponding to about 1.7 GB of memory used.  At higher values, I get allocation errors, though there is about twice that amount of RAM available.
>
> This appears to be a compiler limitation :  on http://support.intel.com/support/performancetools/fortran/linux/sb/CS-007795.htm we read that for IA32 the ifc does not allow arrays larger than 2 GB.
> Documentation on the PGI website suggests the same thing will happen.
>
>
> How do you work around this?
> Thank you very much,
>
> Kevin.
>
>
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>


                                      P.Blaha
--------------------------------------------------------------------------
Peter BLAHA, Inst.f. Materials Chemistry, TU Vienna, A-1060 Vienna
Phone: +43-1-58801-15671             FAX: +43-1-58801-15698
Email: blaha at theochem.tuwien.ac.at    WWW: http://info.tuwien.ac.at/theochem/
--------------------------------------------------------------------------