[gpaw-users] Installation question. Lapack and scalapack.

[Ask Hjorth Larsen]

Hi,

2017-11-20 14:25 GMT+01:00 Gaël Donval via gpaw-users
<gpaw-users at listserv.fysik.dtu.dk>:
(...)
>  * Scalapack is harmful (performance-wise) for small systems and is
>    definitely unnecessary if you don't use "many" CPU cores. So you may
>    want it but you might not actually need it.
(...)

Minor clarification of Gaël's very good explanation: Compiling GPAW
with Scalapack is not harmful to performance in any way.  Using
Scalapack for a too small system will cause overhead, but is unlikely
to cause noticeable degradation of performance in practice because it
affects only a very small part of the calculation.  Nevertheless the
'clean' thing is to only use Scalapack for systems with at least
couple dozens of atoms (LCAO mode) or maybe 100 atoms (FD mode).

Best regards
Ask