[DFTB-Plus-User] Confusion regarding the installation of DFTB+ with Intel compilers

[Bálint Aradi]

Dear Anupriya,

The issues may come from the threading. You have probably way more 
threads on your node, than your problem would scale up to. Intel 
compiled binaries with the MKL-library tend to use all threads, unless 
you tell them differently. GNU-compiled binaries with OpenBlas tend to 
use one thread unless you tell them differently.

When you compare run times between binaries, make sure to set the 
environment variable OMP_NUM_THREADS to the same number.

A few other thoughts:

- We don't suport intel 18 any more, please try to use at least intel 19 
when compiling DFTB+

- The DftbPlusUtils.cmake file is useless for judging what you did, as 
it does not contain any compiler related settings.

- You should read the CMake configuration output, where DFTB+ reports 
which toolchain file it used (e.g. intel.cmake, gnu.cmake, etc.) and 
which compiler flags it used.

Best regards,

Bálint


On 17.11.22 04:44, Anupriya Nyayban wrote:
> 
> Dear DFTB users and experts,
> 
> I am a new user of DFTB+. We have installed dftb 22.1 with Intel 
> compilers (2018 version) and also with gnu compilers separately in a 
> cluster. I could observe the running time of geometry optimizations for 
> Au surface (108 atoms) are 327 sec and 938 sec using dftb+ compiled with 
> Gnu and Intel ones, respectively (with 8 processors). I was expecting 
> the Intel one to be much faster than the gnu one! I am aware of the fact 
> that dftb+ users have already been running for quite a large number of 
> atoms and my aim also is to do so.
> 1) I am unable to trace the possible reasons behind this larger run time 
> for both the cases.
> 2) What could be the reason behind the larger run time with dftb+ 
> compiled with Intel than that with gnu one.
> 3) Please guide me how can I make the best dftb+ executables in order to 
> compute the same system efficiently.
> I have attached the details of the cluster (cpudetails.txt) system, 
> cmake files for dftb+ compiled with Intel compilers 
> (DftbPlusUtils.cmake) herewith.
> 
> If any further information is required please let me know. Eagerly 
> waiting for your help!
> 
> Thank you!
> 
> _______________________________________________
> DFTB-Plus-User mailing list
> DFTB-Plus-User at mailman.zfn.uni-bremen.de
> https://mailman.zfn.uni-bremen.de/cgi-bin/mailman/listinfo/dftb-plus-user

-- 
Dr. Bálint Aradi
Bremen Center for Computational Materials Science, University of Bremen
http://www.bccms.uni-bremen.de/cms/people/b-aradi/


-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature
Type: application/pgp-signature
Size: 840 bytes
Desc: OpenPGP digital signature
URL: <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20221117/1f1c0033/attachment.sig>