[Q-e-developers] QE 6.1 vs. 6.2 performance; Possible starting_charge issue

[Paolo Giannozzi]

On Sat, Oct 28, 2017 at 11:59 AM, Andrii Shyichuk <andrii.shyichuk at chem.uni.
wroc.pl> wrote:

However, QE 6.2 requests ~8.5 Gb memory, while QE 6.1 requests only ~3.5
>

it's an estimate, not a measure. I am not aware of any changes that may
lead QE 6.2 to use more memory than QE 6.1 (except maybe for hybrids) but I
am quite sure that the estimate of QE 6.2 is more realistic than the one of
QE 6.1.


> SCF steps last longer in QE 6.2


this is likely a consequence of the new FFT's, that have better scaling on
many (MPI) processes but are slower on few processors.
I would try MPI parallelization instead of OpenMP, by the way.

, and if I specify starting_charge, the number of steps increases from 9 to
> 23.
>

then don't use it. I don't know what it is good for.

Paolo


> Both codes run on the same supercomputer, were compiled with the same
> libraries and compilers, and both run in MPI+openMP setup, 1 node, 8 cores.
>
> There is another potential issue I'd like to report.
> In the cell of 32 Lu and 48 O, starting charge is calculated as 544 ( =
> 32*8 (Lu3+) + 48*6 (O zero) ), and renormalized to 640 (Lu3+, O2-). In QE
> 6.2, starting_charge can be specified (and is read correctly, according to
> the output file), but that results in starting total charge of 570
> (LO2fc_62), which is renormalized to 640. This is not a problem as is -
> given the renormalization, but might indicate some problem in the routines
> related to starting_charge keyword.
>
> Also, my Lu2O3 calculation shows large negative rho (~3), compared to 640
> electrons in the system, which does not go away at the end of SCF cycle,
> and is not affected by PW cutoff up to 100 Ry and charge density cutoff up
> to 1000 Ry. SCF if achieved anyway.
> Should I ignore this issue?
>
> The Lu potential (neither published nor tested) was obtained from Kevin
> Garrity, but I trust him on this one and will do some testing.
>
> Thank you in advance for your comments.
> Best regards.
> Andrii
>
>
>
>
> _______________________________________________
> Q-e-developers mailing list
> Q-e-developers at qe-forge.org
> http://qe-forge.org/mailman/listinfo/q-e-developers
>
>


-- 
Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche,
Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
Phone +39-0432-558216 <+39%200432%20558216>, fax +39-0432-558222
<+39%200432%20558222>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/developers/attachments/20171028/c204b885/attachment.html>