[Pw_forum] Na3Bi calculation failure
Kane O'Donnell
kane.odonnell at gmail.com
Wed Oct 21 15:09:11 CEST 2015
Works with 5.2.0 (Cray PrgEnv-gnu, fortran 4.9, Cray libsci) with -ndiag > 1, -ntg > 1. Thanks! Lesson learned…
Kane
> On 21 Oct 2015, at 17:27, Paolo Giannozzi <p.giannozzi at gmail.com> wrote:
>
> Unless you need new developments that are available in the svn version only, please try if it works with the 5.2.0 version. We just found a problem (also affecting v.5.2.1) with "task groups" that may lead to strange crashes.
>
> Paolo
>
> On Wed, Oct 21, 2015 at 11:04 AM, Kane O'Donnell <kane.odonnell at gmail.com <mailto:kane.odonnell at gmail.com>> wrote:
>
> Hi all,
>
> Wondering if I can get some help trying to diagnose a crash. I’m running the SVN latest on a Cray XC40 (Magnus - https://www.pawsey.org.au/our-systems/magnus-technical-specifications/ <https://www.pawsey.org.au/our-systems/magnus-technical-specifications/>). Usually no problems, but I have difficulties getting the attached slab calculation to run past the first few davidson diagonalizations. It’s a 3x3 c-oriented slab of Na3Bi, the minimum I can use to capture a certain adsorbate reconstruction (this just the bare slab). It’s only a 72 atom cell but Bi has a lot of electrons (I think there’s about ~800 electrons and ~900 bands). I have spin-orbit coupling switched on (important for this solid), and I have been able to do calculations on the smaller unit cell using the library pseudopotentials listed in the species block. Calculations on systems of this size (e.g. O(1000) electrons, bands) are routine on Magnus, so I think I’m probably just doing something stupid but can’t seem to figure it out.
>
> Typical run conditions are with 384 processors (16 nodes, 24 cores), with -nk 3 -ndiag 100 -ntg 8. Moving down to ~12 nodes leads to a out of memory crash just as the code reports it is allocating random wf’s at the beginning. From 16 nodes and upwards, the crashes happen around the diagonalization step. Switching to CG works but the slowdown is astronomical (~10000 seconds per SCF step, not feasible for a relaxation). A typical output is attached. With -ndiag > 1, the error is “problems computing cholesky”, with -ndiag = 1, the error is "S matrix not positive definite”, both from cdiaghg. A search of the forums suggests this issue comes up every now and then on wildly different systems and is usually blamed on the user/compiler/lapack/blas/scalapack. So, details: QE was compiled by me on Magnus with PrgEnv-gnu (fortran 4.9.0) against the Cray libsci (includes fftw, scalapack, etc), with:
>
> ./configure —enable-parallel —with-scalapack=yes FC=ftn CC=cc
>
> and all tests are passed with no problems.
>
> Any ideas? Let me know if there is any further information necessary.
>
> Best regards,
>
> Kane
> Kane O'Donnell
> Postdoctoral Research Fellow | Department of Physics, Astronomy and Medical Radiation Science
>
> Curtin University
> Tel | +61 8 9266 1381 <tel:%2B61%208%209266%201381>
> Fax | +61 8 9266 2377 <tel:%2B61%208%209266%202377>
>
> Email | kane.odonnell at curtin.edu.au <>
>
>
>
> Curtin University is a trademark of Curtin University of Technology
> CRICOS Provider Code 00301J
>
> &control
> calculation = 'relax',
> title = '',
> outdir = './',
> prefix = 'Na3Bi_331',
> pseudo_dir = '/group/partner1197/kodonnell/qe_pseudos/PSEUDOPOTENTIALS/',
> wf_collect = .true.
> /
> &system
> ibrav = 0,
> nat = 72,
> ntyp = 2,
> nbnd = 896,
> ecutwfc = 50,
> !ecutrho = 280,
> !tot_charge=+1.0,
> occupations = 'smearing',
> smearing = 'mv',
> degauss = 0.0073,
> lspinorb = .true.,
> noncolin = .true.,
> starting_magnetization(1) = 0.0,
> starting_magnetization(2) = 0.0
> /
> &electrons
> conv_thr = 1.0D-7
> /
> &ions
> /
> ATOMIC_SPECIES
> Bi 1.0 Bi.rel-pbe-dn-kjpaw_psl.0.2.2.UPF
> Na 1.0 Na.rel-pbe-spn-kjpaw_psl.0.2.UPF
> CELL_PARAMETERS angstrom
> 16.344 0 0
> -8.172 14.1543 0
> 1.77359e-15 3.07196e-15 28.965
> K_POINTS automatic
> 2 2 1 0 0 0
> ATOMIC_POSITIONS angstrom
> Bi 0 3.146 2.414 0 0 0
> Na 2.724 4.718 2.414 0 0 0
> Na 0 3.146 5.629
> Bi 2.724 1.573 7.241
> Na 2.724 4.718 7.241
> Na 2.724 1.573 0.801 0 0 0
> Na 2.724 1.573 4.026
> Na 0 3.146 8.854
> Bi -2.724 7.864 2.414 0 0 0
> Na 0 9.436 2.414 0 0 0
> Na -2.724 7.864 5.629
> Bi 0 6.291 7.241
> Na 0 9.436 7.241
> Na 0 6.291 0.801 0 0 0
> Na 0 6.291 4.026
> Na -2.724 7.864 8.854
> Bi -5.448 12.582 2.414 0 0 0
> Na -2.724 14.154 2.414 0 0 0
> Na -5.448 12.582 5.629
> Bi -2.724 11.009 7.241
> Na -2.724 14.154 7.241
> Na -2.724 11.009 0.801 0 0 0
> Na -2.724 11.009 4.026
> Na -5.448 12.582 8.854
> Bi 5.448 3.146 2.414 0 0 0
> Na 8.172 4.718 2.414 0 0 0
> Na 5.448 3.146 5.629
> Bi 8.172 1.573 7.241
> Na 8.172 4.718 7.241
> Na 8.172 1.573 0.801 0 0 0
> Na 8.172 1.573 4.026
> Na 5.448 3.146 8.854
> Bi 2.724 7.864 2.414 0 0 0
> Na 5.448 9.436 2.414 0 0 0
> Na 2.724 7.864 5.629
> Bi 5.448 6.291 7.241
> Na 5.448 9.436 7.241
> Na 5.448 6.291 0.801 0 0 0
> Na 5.448 6.291 4.026
> Na 2.724 7.864 8.854
> Bi 0 12.582 2.414 0 0 0
> Na 2.724 14.154 2.414 0 0 0
> Na 0 12.582 5.629
> Bi 2.724 11.009 7.241
> Na 2.724 14.154 7.241
> Na 2.724 11.009 0.801 0 0 0
> Na 2.724 11.009 4.026
> Na 0 12.582 8.854
> Bi 10.896 3.146 2.414 0 0 0
> Na 13.62 4.718 2.414 0 0 0
> Na 10.896 3.146 5.629
> Bi 13.62 1.573 7.241
> Na 13.62 4.718 7.241
> Na 13.62 1.573 0.801 0 0 0
> Na 13.62 1.573 4.026
> Na 10.896 3.146 8.854
> Bi 8.172 7.864 2.414 0 0 0
> Na 10.896 9.436 2.414 0 0 0
> Na 8.172 7.864 5.629
> Bi 10.896 6.291 7.241
> Na 10.896 9.436 7.241
> Na 10.896 6.291 0.801 0 0 0
> Na 10.896 6.291 4.026
> Na 8.172 7.864 8.854
> Bi 5.448 12.582 2.414 0 0 0
> Na 8.172 14.154 2.414 0 0 0
> Na 5.448 12.582 5.629
> Bi 8.172 11.009 7.241
> Na 8.172 14.154 7.241
> Na 8.172 11.009 0.801 0 0 0
> Na 8.172 11.009 4.026
> Na 5.448 12.582 8.854
>
>
>
>
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org <mailto:Pw_forum at pwscf.org>
> http://pwscf.org/mailman/listinfo/pw_forum <http://pwscf.org/mailman/listinfo/pw_forum>
>
>
>
> --
> Paolo Giannozzi, Dept. Chemistry&Physics&Environment,
> Univ. Udine, via delle Scienze 208, 33100 Udine, Italy
> Phone +39-0432-558216 <tel:%2B39-0432-558216>, fax +39-0432-558222 <tel:%2B39-0432-558222>_______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org <mailto:Pw_forum at pwscf.org>
> http://pwscf.org/mailman/listinfo/pw_forum <http://pwscf.org/mailman/listinfo/pw_forum>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20151021/123a179f/attachment.html>
More information about the users
mailing list