[gpaw-users] Computational cost of vdW-DF

[Duy Le]

On Tue, Apr 12, 2011 at 12:52 PM,  <andreasm at fysik.dtu.dk> wrote:
> I guess you can change the number of kernels to be consistent with the
> number of cpus
> For example using 16 cpus:
> vdw=FFTVDWFunctional(name='vdW-DF2',Nalpha=16)
> Than everything above 16 cpus will only help the GGA part of the
> calculations.

By defaults, it is 20. Don't see how it helps if we reduce it to 16.

> I have a question in this respect. What would take to change the
> implementation such that you could use more cpus? It would be very helpful
> if one could use 32 or even 64...

--------------------------------------------------
Duy Le
PhD Student
Department of Physics
University of Central Florida.

"Men don't need hand to do things"
>
> Best
> Andreas
>
>> 2011/4/12 Jens Jørgen Mortensen <jensj at fysik.dtu.dk>:
>>> On Tue, 2011-04-12 at 15:29 +0200, Duy Le wrote:
>>>> 2011/4/12 Jens Jørgen Mortensen <jensj at fysik.dtu.dk>:
>>>> > On Mon, 2011-04-11 at 23:56 +0200, Duy Le wrote:
>>>> >> Dear all,
>>>> >> As I understand, the approach of Román-Pérez and Soler (PRL 2009)
>>>> has
>>>> >> been implemented in GPAW.  I wonder why the self-consistent vdW-DF
>>>> >> calculation is still extremely expensive (3 or 4 times in most of my
>>>> >> calculations) in comparison to regular GGA-PBE calculation.
>>>> >> Does any of you experience this slowness? I would like to know the
>>>> reason.
>>>> >
>>>> > For a small system with few electrons, the calculation of the vdW-DF
>>>> > potential will dominate.  With more electrons the vdW-DF part should
>>>> > become smaller unless you run on so many cores that the
>>>> parallelization
>>>> > of the vdW-DF stuff stops scaling.  So, what is your system and how
>>>> many
>>>> > cpu's do you use?
>>>> It is the case, my systems are big, in the order of about 70-150
>>>> atoms. Of course I have to use lot of cores (64 to 512 cores in most
>>>> of the case) otherwise I don't have enough memory allocated.
>>>> Is there any way to improve it? The PBE calculation is scaled very
>>>> well.
>>>
>>> The current implementation of the vdW-DF xc-potential scales only up to
>>> 20 cores.  Could you show us the text output from one of your
>>> calculations - you should be able to see the timings for the different
>>> parts of the calculation in the text output.
>>>
>> Sure. Check attachments for a scf PBE and vdW-DF1. I deleted many
>> lines which are not so important.
>> Those are done with 64 cores on AIX power6.
>> --------------------------------------------------
>> Duy Le
>> PhD Student
>> Department of Physics
>> University of Central Florida.
>>
>> "Men don't need hand to do things"
>>> Jens Jørgen
>>>
>>>> > Jens Jørgen
>>>> >
>>>>
>>>> --------------------------------------------------
>>>> Duy Le
>>>> PhD Student
>>>> Department of Physics
>>>> University of Central Florida.
>>>>
>>>> "Men don't need hand to do things"
>>>> >> Thank you.
>>>> >> --------------------------------------------------
>>>> >> Duy Le
>>>> >> PhD Student
>>>> >> Department of Physics
>>>> >> University of Central Florida.
>>>> >>
>>>> >> "Men don't need hand to do things"
>>>> >>
>>>> >> _______________________________________________
>>>> >> gpaw-users mailing list
>>>> >> gpaw-users at listserv.fysik.dtu.dk
>>>> >> https://listserv.fysik.dtu.dk/mailman/listinfo/gpaw-users
>>>> >
>>>> >
>>>
>>>
>> _______________________________________________
>> gpaw-users mailing list
>> gpaw-users at listserv.fysik.dtu.dk
>> https://listserv.fysik.dtu.dk/mailman/listinfo/gpaw-users
>
>
>