[QE-users] High k-point density memory estimates

[Paolo Giannozzi]

The memory usage is very hard to estimate in general. Morever it depends
upon a number of factors. Old versions for instance used to keep by default
a single set of wavefunctions for one k-point, store those for the other
k-points to disk. Since some time the default is to keep wavefunctions for
all k-points in memory.  Recent versions of QE have a better estimate of
how much memory is allocated.

Paolo

On Tue, Oct 22, 2019 at 6:10 PM Eric Glen Suter <esuter at uga.edu> wrote:

> Hi all,
>
> I'm running jobs with a relatively high number of kpoints with QE 6.2. For
> a while I was using the dynamical memory estimates output by the pw.x
> executable as a guidepost for requesting memory on a cluster, and this
> worked well when I had a relatively small number of kpoints. But it turns
> out that for these dense grids I need significantly more than the estimate
> indicates. The estimate says a few hundred Mb, but the actual resources
> used indicate it's more on the order of several tens of Gb.
>
> I find in the user manual an estimate of the number of double precision
> complex floating point numbers that would be needed (
>
> O = mMN + PN + pN1N2N3 + qNr1Nr2Nr3
> ) and this seems to give an estimate on the order of what the output file
> says.
>
> Is there something I'm missing that goes into determining how much memory
> a job should take? I'm mostly interested to know if there's a good way to
> predict how much memory future jobs will need so I can make a more educated
> guess when I request memory.
>
> Any insights you can offer are greatly appreciated.
>
> Best regards,
>
> Eric Suter
>
> ----------------------------
>
> PhD Candidate, Dept. of Physics and Astronomy
>
> Center for Simulational Physics
>
> University of Georgia
>
> ----------------------------
>
> email: esuter at uga.edu
>
> phone: 912-856-3071
>
> _______________________________________________
> Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso)
> users mailing list users at lists.quantum-espresso.org
> https://lists.quantum-espresso.org/mailman/listinfo/users



-- 
Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche,
Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
Phone +39-0432-558216, fax +39-0432-558222
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20191022/7d904aa2/attachment.html>