[Pw_forum] parallel efficiency for large system

Davide Ceresoli davide.ceresoli at materials.ox.ac.uk
Wed Jan 26 17:01:20 CET 2011


On 01/22/2011 08:52 AM, Paolo Giannozzi wrote:
>
> On Jan 21, 2011, at 21:11 , Yuyang Zhang wrote:
>
>> try to use -ntg tag when submitting the jobs but no significant
>> improvement.
>
> note that versions up to 4.2.1 have a bug in the spin-polarized case
> at Gamma point if you use -ntg
>

I'm very interested in this thread. Is there a follow-up? Are there
any news?

I can add to the discussion my experience on a ~500 atoms, insulating
systems (~2200 electrons): using CPV, I've managed to relax the system
by damped dynamics (electrons and ions) on 32 cores in 5 days (1 day
to get wavefunctions, 4 days to relax).

Instead PW took 5 days to complete the first SCF (i.e. to obtain
ground state wavefunctions and forces), and therefore I killed the job.
I can't tell how many ionic steps it would need to complete.

If your system a well-behaving insulator, CPV is definitely the way
to go.

Cheers,
     Davide



-- 
+----------------------------------------------------------+
    Davide Ceresoli <davide.ceresoli at materials.ox.ac.uk>
    University of Oxford - Department of Materials
    Parks Road, Oxford OX1 3PH, United Kingdom
    Phone: +44-1865-612793
    Mobile:+44-7767-091256, +39-347-1001570
    Skype: dceresoli
+----------------------------------------------------------+



More information about the users mailing list