[gpaw-users] Gpaw on taito (CSC)

[Marcin Dulak]

On 07/25/2014 03:26 PM, Marcin Dulak wrote:
> On 07/25/2014 02:04 PM, leonardo espinosa wrote:
>> Hi,
>>
>> I've have tested different combinations of optimization algorithms 
>> with the option that
>> Marcin suggested: export OMP_NUM_THREADS=1, and different 
>> combinations of versions of ase/gpaw.
>> The calculation still crashes with the same error message. I tried 
>> other options such us:
>>
>> export NUMEXPR_NUM_THREADS=1
>> export MKL_NUM_THREADS=1
>> export MKL_DOMAIN_NUM_THREADS=1
>> export OMP_NUM_THREADS=1
>> export OPENBLAS_NUM_THREADS=1
>> export GOTO_NUM_THREADS=1
>>
>> But nothing, the same error message.
>>
>> Maybe I should start from the scratch, installing my own version.
>> However I though that somebody, gpaw user in TAITO can give
>> me a hand with a script o some hints.
>>
> when you run tests in parallel (e.g. on 8 cores), do any fail?
> mpiexec gpaw-python `which gpaw-test`
please see also this comment (are you using scalapack?):
http://listserv.fysik.dtu.dk/pipermail/gpaw-users/2014-January/002523.html

Best regards,

Marcin
>
> Best regards,
>
> Marcin
>
>> Thanks again and any help with this annoying bug is always welcome.
>>
>> Best,
>>
>> leo
>>
>>
>>
>> ---
>> Leonardo A. Espinosa L.
>> Postdoctoral Researcher,
>> Aalto University,
>> Espoo, Finland.
>> http://goo.gl/WjBIXv <http://yonk.is/gI3>
>>
>>
>>
>>
>> On Thu, Jul 24, 2014 at 5:33 PM, Marcin Dulak 
>> <Marcin.Dulak at fysik.dtu.dk <mailto:Marcin.Dulak at fysik.dtu.dk>> wrote:
>>
>>     On 07/24/2014 08:39 AM, Jens Jørgen Mortensen wrote:
>>>     On 07/23/2014 04:41 PM, leonardo espinosa wrote:
>>>>     Hi Ask,
>>>>
>>>>     Here I attached you the files,
>>>>     thanks for your help.
>>>
>>>     It looks like the positions are out of sync on the different
>>>     cores.  I think GPAW should return identical energies and forces
>>>     from all cores.  Maybe the BFGS algorithm is causing this?  If
>>>     that is the case then how do we deal with this?
>>     is GPAW or numpy linked against a multithreaded blas?
>>     From
>>     https://wiki.fysik.dtu.dk/gpaw/install/installationguide.html -
>>     "Multithreading is not supported."
>>     For some libraries like mkl,openblas you can disable
>>     multithreading with: export OMP_NUM_THREADS=1
>>
>>     Best regards,
>>
>>     Marcin
>>
>>
>>>
>>>     Maybe GPAW should accept a small divergence in the positions due
>>>     to numerical noise and then just use the positions from rank 0?
>>>
>>>     Jens Jørgen
>>>
>>>>     leo
>>>>
>>>>     ---
>>>>     Leonardo A. Espinosa L.
>>>>     Postdoctoral Researcher,
>>>>     Aalto University,
>>>>     Espoo, Finland.
>>>>     http://goo.gl/WjBIXv <http://yonk.is/gI3>
>>>>
>>>>
>>>>
>>>>
>>>>     On Wed, Jul 23, 2014 at 5:26 PM, Ask Hjorth Larsen
>>>>     <asklarsen at gmail.com <mailto:asklarsen at gmail.com>> wrote:
>>>>
>>>>         Hi Leonardo
>>>>
>>>>         What is the error?  Also the logfile.
>>>>
>>>>         Best regards
>>>>         Ask
>>>>
>>>>         2014-07-23 16:23 GMT+02:00 leonardo espinosa
>>>>         <espinosa.leal at gmail.com <mailto:espinosa.leal at gmail.com>>:
>>>>         > Dear gpaw users,
>>>>         >
>>>>         > I'm trying to run a simple BFGS optimization of a benzene
>>>>         > molecule on taito cluster
>>>>         (https://research.csc.fi/taito-user-guide),
>>>>         > using different versions of gpaw: 0.9.0 and 0.10.0. I'm
>>>>         having
>>>>         > some problems because after certain number of iterations the
>>>>         > calculation crashes, anybody here that use gpaw-0.10.0 on
>>>>         taito
>>>>         > without problem? Can you supply me with an example that runs?
>>>>         >
>>>>         > I attached you here the files that I'm using.
>>>>         >
>>>>         > Best regards,
>>>>         >
>>>>         > leo
>>>>         >
>>>>         >
>>>>         >
>>>>         >
>>>>         > ---
>>>>         > Leonardo A. Espinosa L.
>>>>         > Postdoctoral Researcher,
>>>>         > Aalto University,
>>>>         > Espoo, Finland.
>>>>         > http://goo.gl/WjBIXv
>>>>         >
>>>>         >
>>>>         >
>>>>
>


-- 
***********************************
  
Marcin Dulak
Technical University of Denmark
Department of Physics
Building 307, Room 229
DK-2800 Kongens Lyngby
Denmark
Tel.: (+45) 4525 3157
Fax.: (+45) 4593 2399
email: Marcin.Dulak at fysik.dtu.dk

***********************************

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.fysik.dtu.dk/pipermail/gpaw-users/attachments/20140725/dfff43fe/attachment-0001.html>