[Pw_forum] espresso 2.1.4 and ifc8.1.024
Nicola Marzari
marzari at MIT.EDU
Thu Jul 14 17:15:00 CEST 2005
Thanks Patricia !
There is some mistery going around here - do make sure hat the
three calculations with pwscf 2.0.4 are all taking the same number of
bfgs steps, and the same number of iterations on the electrons (by the
way, that could explain the difference coming just from different
smearings).
Ideally, you want to have tests that always have the same number of
iterations - i.e. I would avoid ionic relaxations, and would do a
fixed number of electronic iterations (i.e., use a very tight electronic
threshold, but limit the number of iterations to, say, 10, so that the
code doesn't take less than 10 to converge).
The most relevant issue seem to be the huge difference between 2.0.4 and
2.1.1 - could you redo that on one single node, making sure noone else
run on the machine ?
In a parallel job, and with a shared machine, all timings can be really
thrown off.
Thanks,
nicola
Patricia Paredes wrote:
> Hi Nicola - Paolo:
> OK, now I made all the test (including marzari-vanderbilt smearing...)
> and changing the pseupotentials from the old format
> to the *.UPF format currently on the pwscf web page.
> here are the results for a cluster of PIV-3.0GHZ, compiled with ifc
> 8.0.034 (both pw and the mpi) running on 5 nodes.
>
>
> pwscf 2.0.4:
> bfgs converged in 10 scf cycles and 8 bfgs steps
> final energy: -1259.5907420114 ryd
>
>
> electrons : 8813.48s CPU ( 10 calls, 881.348 s avg)
> first ionic iteration
> scf - last iteration: 1527.75 secs
> scf + forces : 1674.88 secs
>
> PWSCF : 2h35m CPU time
>
> using smearing='marzari-vanderbilt'
> PWSCF : 2h 0m CPU time
>
> using pseudopotential in UPF format+ marzari-vanderbilt smearing
> S 32.06 S.pbe-van_bm.UPF
> C 12.01 C.pbe-van_bm.UPF
> H 1 H.pbe-van_bm.UPF
> Au 196.96 Au.pbe-nd-van.UPF
>
> PWSCF : 1h18m cpu time!!!!
>
>
> pwscf 2.1.1
> bfgs converged in 12 scf cycles and 11 bfgs steps
> Final energy = -1259.5907145126 ryd
>
> electrons : 11685.19s CPU ( 12 calls, 973.766 s avg)
> fist ionic iteration
> scf - last iteration: 1539.16 secs
> scf + forces : 1685.91 secs
>
>
> PWSCF : 3h24m CPU time
>
> using smearing='marzari-vanderbilt'
> PWSCF : 3h33m CPU time
>
> using pseudopotential in UPF format+ marzari-vanderbilt smearing
> S 32.06 S.pbe-van_bm.UPF
> C 12.01 C.pbe-van_bm.UPF
> H 1 H.pbe-van_bm.UPF
> Au 196.96 Au.pbe-nd-van.UPF
>
> PWSCF : 3h30m CPU time
>
> pwscf 2.1.2
> bfgs converged in 12 scf cycles and 11 bfgs steps
> Final energy = -1259.5907141926 ryd
>
> electrons : 13977.43s CPU ( 12 calls,1164.786 s avg)
> fist ionic iteration
> scf - last iteration:2028.56 secs
> scf + forces : 2181.71 secs
>
>
>
>
>
> A 10:30 AM 7/10/2005 -0400, ha escrito:
> >
> >
> >Thanks a lot Patricia - these are *exceedingly* useful tests.
> >
> >We have even a standard test for CP:
> >http://www.democritos.it/pipermail/pw_forum/2004-October/001483.html
> >
> >Axel had kindly provided some results, still in the repository, and
> >that I should add to the report.
> >
> >If you'd like, we can add yours as a standard test for PWSCF. Of course,
> >you should switch to a "marzari-vanderbilt" smearing, then :-) .
> >
> > nicola
> >
> >
> >Patricia Paredes wrote:
> >
> >> Hi Axel,
> >> I made some experiments compiling all the versions of pwscf since 2.0.3
> >> using ifc8.1.024, cluster mkl 7.1 and compiling mpi with the same
> compiler.
> >> first iteration
> >> 2.0.3 117.11 secs
> >> 2.0.4 109.89 secs
> >> 2.1.1 110.03 secs
> >> 2.1.2 164.93 secs
> >> 2.1.3 152.52 secs
> >> 2.1.4 154.43 secs
> >>
> >> So, the breaking point was version 2.1.1 (btw, v2.1.0 didn´t work). In
> >> the web page it is stated that it was a modification in soubroutine
> >> electrons.f90 to make a double check... (By carlo sbraccia), perhaps
> >> there is the point: the double check. I was taking a look to the final
> >> statistics and the major differences were at electrons.f90 (Sorry, I
> >> don´t have the numbers now since I overwrite the files :).
> >>
> >> The input file was:
> >>
> >> cuatricapa de Au
> >> Au
> >> &control
> >> calculation='relax'
> >> restart_mode='from_scratch',
> >> prefix='prueba',
> >> pseudo_dir = '/home/pwscf/pseudo/',
> >> outdir='/home/patricia/plata/temp/'
> >> /
> >> &system
> >> ibrav=0,
> >> nat=17,
> >> ntyp=4,
> >> ecutwfc = 22.0,
> >> ecutrho = 180.0
> >> celldm(1)=7.89304630
> >> degauss=0.03
> >> occupations='smearing'
> >> smearing='methfessel-paxton'
> >> /
> >> &electrons
> >> mixing_mode = 'local-TF'
> >> conv_thr = 1.0d-6
> >> mixing_beta = 0.5
> >> diagonalization = 'cg'
> >> /
> >> &ions
> >> ion_dynamics = 'bfgs'
> >> /
> >> ATOMIC_SPECIES
> >> S 32.06 S.gpbe.vdb
> >> C 12.01 C.gpbe.vdb
> >> H 1 H.gpbe.vdb
> >> Au 196.96 Au.pbe-nd-van.UPF
> >> ATOMIC_POSITIONS {alat}
> >> H -0.187034332 0.221830025 4.921552258
> >> H -0.013559323 0.486812870 4.628026988
> >> H 0.242342701 0.256878798 4.888036613
> >> C 0.014538967 0.257636565 4.754725119
> >> S 0.016846606 -0.082426513 4.477818864
> >> Au 0.383806249 0.011800463 4.017799873
> >> Au 1.076105787 -0.011642255 3.950466565
> >> Au 1.768205765 0.011736229 4.018611550
> >> Au 0.706660429 -0.205310574 3.426597457
> >> Au 1.418423038 -0.205689922 3.424408560
> >> Au 1.062125184 0.398360581 3.405678032
> >> Au 0.701778921 0.206802972 2.843998759
> >> Au 1.435470249 0.207226266 2.845135398
> >> Au 1.067569501 -0.420576720 2.851998036
> >> Au 0.353553300 0.000000000 2.267949200 0 0 0
> >> Au 1.060660200 0.000000000 2.267949200 0 0 0
> >> Au 1.767767000 0.000000000 2.267949200 0 0 0
> >> CELL_PARAMETERS {cubic}
> >> 1.0606602 0.6123724 0.0
> >> 1.0606602 -0.6123724 0.0
> >> 0.0000000 0.0 7.0
> >> K_POINTS {automatic}
> >> 4 4 1 0 0 0
> >>
> >>
> >> *_Prof. Dr. Patricia Paredes Olivera
> >> _*Facultad de Ciencias Químicas
> >> Universidad Nacional de Córdoba
> >> ph/fax: 054-0351-434-4972
> >> _______________________________________________ Pw_forum mailing list
> >> Pw_forum at pwscf.org http://www.democritos.it/mailman/listinfo/pw_forum
> >
> >--
> >---------------------------------------------------------------------
> >Prof Nicola Marzari Department of Materials Science and Engineering
> >13-5066 MIT 77 Massachusetts Avenue Cambridge MA 02139-4307 USA
> >tel 617.4522758 fax 617.2586534 marzari at mit.edu http://nnn.mit.edu
> >_______________________________________________
> >Pw_forum mailing list
> >Pw_forum at pwscf.org
> >http://www.democritos.it/mailman/listinfo/pw_forum
> >
> *_Prof. Dr. Patricia Paredes Olivera
> _*Facultad de Ciencias Químicas
> Universidad Nacional de Córdoba
> ph/fax: 054-0351-434-4972
> _______________________________________________ Pw_forum mailing list
> Pw_forum at pwscf.org http://www.democritos.it/mailman/listinfo/pw_forum
--
---------------------------------------------------------------------
Prof Nicola Marzari Department of Materials Science and Engineering
13-5066 MIT 77 Massachusetts Avenue Cambridge MA 02139-4307 USA
tel 617.4522758 fax 617.2586534 marzari at mit.edu http://nnn.mit.edu
More information about the users
mailing list