Page 1 of 1

standard benchmark fails

Posted: Wed Nov 04, 2009 5:18 pm
by cubestack
Dear users and developers,

compiling vasp4_6 worked without problems on a quad core Intel system. In order to prepare for a bigger installation with mpi the actual installation is only serial using the intel mkl for blas and lapack, and leaving the rest as it comes out of the box.
Compiler: ifort/10.1.015
IntelMKL: imkl/10.0.3.020
in order to check the installation the bench and bench.Hg were chosen. Unfortunately both tests fail with messages:
...
very serious problems the old and the new charge density differ
...
Trying different compiler switches and optimization levels does not alter this effect.
Also setting IALGO=48 in bench as described elswhere does not help.
A successful compilation with failing standard tests is not really satisfactory, so a hint to get a first unoptimized installation properly would be very helpful.

standard benchmark fails

Posted: Tue Nov 10, 2009 5:27 pm
by cubestack
Output from bench.Hg looks the following:

######## snip #######

entering main loop
N E dE d eps ncg rms rms(c)
RMM: 1 -0.634645362635E+05 -0.63465E+05 -0.25341E+05 316 0.973E+02
RMM: 2 -0.573875314051E+05 0.60770E+04 -0.32406E+04 316 0.350E+02
RMM: 3 -0.578139897883E+05 -0.42646E+03 -0.91175E+03 316 0.260E+02
RMM: 4 -0.577731187268E+05 0.40871E+02 0.38934E+02 316 0.225E+02
RMM: 5 -0.579321778141E+05 -0.15906E+03 -0.14070E+03 316 0.244E+02
RMM: 6 -0.587361191535E+05 -0.80394E+03 -0.10712E+04 1196 0.276E+02
RMM: 7 -0.587636750986E+05 -0.27556E+02 -0.43535E+03 1120 0.167E+02BRMIX: very serious problems
the old and the new charge density differ
old charge density: 573.26112 new 570.35990
0.214E+02
RMM: 8 -0.491397530256E+05 0.96239E+04 -0.18632E+04 1254 0.253E+02BRMIX: very serious problems
the old and the new charge density differ
old charge density: 568.29427 new 570.57386
0.123E+02

######## snap ##########

standard benchmark fails

Posted: Mon Nov 30, 2009 9:52 am
by admin
1) please check if all libraries are installed correctly, by using EXACTLY the same BLAS and LAPACK with your serial executable.
2) if this is ok, please run the parallel executable on one node only (this is to check whether
-- the errror is due to some communication error (hardware-related)
-- or if the parallel FFT-library is not installed correctly, in case you do NOT use Jürgen Furthmüller's FFT)