[Pw_forum] Benchmarking QE 4.1

[Marcos Veríssimo Alves]

Hi all,

I have recently compiled QE in our cluster using mpich 1.2.7p1 for
myrinet, using the pathscale compiler 2.4. Compilation went smoothly
after the advice given by Lorenzo Paulatto and Paolo Giannozzi
yesterday, and today I started benchmarking it for a small system on
which I will have a lot of k-points (around 1700 because no symmetries
are detected, and it's tetragonal SrTiO3). The time bewteen the end of
the initialization of the calculation (starting mode="from scratch")
and end of the first calculation step are much longer than the typical
times for the subsequent davidson diagonalization steps (between 4-5
times longer):

#k-points   1st Davidson step (s)        typical time for subsequent steps (s)

4                       116                                             22
6                       174                                             34
27                     780                                            139
48                    1355                                           240

Does anyone have any idea of why this happens? It doesn't happen for
our compilation of siesta, for example. Any suggestions as to how to
avoid this? I'll be glad to provide any further info that might be
needed...

Cheers,

Marcos