[Pw_forum] How to speed up gipaw?

Davide Ceresoli davide.ceresoli at materials.ox.ac.uk
Wed Sep 7 11:47:09 CEST 2011


Dear Pengju,
     on possibility is to use pools, in order to distribute k-points on more
CPUs. I always set iverbosity=1 to see some output as the calculation
is running (I suffer of empty-output-phobia).

To get an idea of the computational cost: I have calculated a complicated
silicate mineral with 88 atoms in the unit cell. 4 special k-points. The
calculation took 1 day on 16 CPUs.

Davide

On 01/-10/-28163 08:59 PM, Ren PJ wrote:
> Dear all,
> I'm using gipaw to calculate NMR of a system contain more than 100
> atoms, but the first k point haven't been done after more than 40
> hours. It's so slow for me.
> Here is my input:
> &inputgipaw
>          job =nmr'
>          prefix =pw'
>          tmp_dir =./tmp/'
>          isolve =
>          conv_threshold
-10
>          iverbosity =
>          q_gipaw = .01
>          spline_ps =true.
>          use_nmr_macroscopic_shape =true.
>         nmr_macroscopic_shape =.5 0.0 0.0 0.0 0.5 0.0 0.0 0.0 0.6667
> /
> Can anyone figure me out how to improve this?
> PS. My computer is 8 core cpu, and gipaw and QE is compiled with intel
> mkl lib.
> Thanks very much!
> 
> Pengju Ren
> 
> Dalian Institute of Chemical Physics,
> Chinese Academy of Science
> 



More information about the users mailing list