[QE-users] Huge memory requirements for a scf calculation of a system consisting with 300+ atoms.

Paolo Giannozzi p.giannozzi at gmail.com
Tue May 26 09:23:02 CEST 2020


With 300+ atoms, relatively large cutoffs (60/480), several k-points, yes,
the amount of needed RAM memory is large. If you want to stick to your
parameters (cutoffs and k-points) you may reduce the amount of needed
memory by
- parallelizing over as many pools as k-points (avoids "w. buffer" below)
- setting "diago_david_ndim" to 2
- patch "addusforce" as follows:

diff --git a/PW/src/addusforce.f90 b/PW/src/addusforce.f90
index 3c353493d..d79dd8baa 100644
--- a/PW/src/addusforce.f90
+++ b/PW/src/addusforce.f90
@@ -137,7 +137,7 @@ SUBROUTINE addusforce_g( forcenl )
         DO na = 1, nat
            IF ( ityp(na) == nt ) nab = nab + 1
         ENDDO
-        ALLOCATE ( aux1( ngm_l, na, 3) )
+        ALLOCATE ( aux1( ngm_l, nab, 3) )
         ALLOCATE ( ddeeq(nij, nab, 3, nspin_mag) )
         !
         DO is = 1, nspin_mag

Paolo

On Tue, May 26, 2020 at 5:44 AM Hongyi Zhao <hongyi.zhao at gmail.com> wrote:

> Hi,
>
> I try to run a scf calculation of my system consisting with 300+
> atoms, but the pw.x gives the following estimated resource
> requirements:
>
>      Dynamical RAM for                 wfc:     204.96 MB
>
>      Dynamical RAM for     wfc (w. buffer):    1844.62 MB
>
>      Dynamical RAM for           str. fact:      25.69 MB
>
>      Dynamical RAM for           local pot:       0.00 MB
>
>      Dynamical RAM for          nlocal pot:    1258.70 MB
>
>      Dynamical RAM for                qrad:       7.03 MB
>
>      Dynamical RAM for          rho,v,vnew:      45.97 MB
>
>      Dynamical RAM for               rhoin:      15.32 MB
>
>      Dynamical RAM for            rho*nmix:     102.78 MB
>
>      Dynamical RAM for           G-vectors:      25.22 MB
>
>      Dynamical RAM for          h,s,v(r/c):      10.59 MB
>
>      Dynamical RAM for          <psi|beta>:      48.85 MB
>
>      Dynamical RAM for                 psi:     819.83 MB
>
>      Dynamical RAM for                hpsi:     819.83 MB
>
>      Dynamical RAM for                spsi:     819.83 MB
>
>      Dynamical RAM for      wfcinit/wfcrot:     703.07 MB
>
>      Dynamical RAM for           addusdens:    2479.53 MB
>
>      Dynamical RAM for          addusforce:    7372.39 MB
>
>      Dynamical RAM for         addusstress:    1291.15 MB
>
>      Estimated static dynamical RAM per process >       3.32 GB
>
>      Estimated max dynamical RAM per process >      10.52 GB
>
>      Estimated total dynamical RAM >     462.96 GB
>
>
> And finally the job was killed by the system with the following message:
>
>
> ===================================================================================
> =   BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
> =   RANK 0 PID 404746 RUNNING AT X10DAi-01
> =   KILLED BY SIGNAL: 9 (Killed)
>
> ===================================================================================
>
> The input file and pseudos used by the above example is located here:
> https://github.com/hongyi-zhao/test/raw/master/PWscf.zip
>
> Could someone please try it and confirm whether it's really a so
> resource demanding job?
>
> Regards
> --
> Hongyi Zhao <hongyi.zhao at gmail.com>
> _______________________________________________
> 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/20200526/a7cceb17/attachment.html>


More information about the users mailing list