Page 1 of 1

Different calculation results for different systems

PostPosted: Mon Jan 14, 2019 9:20 pm
by cgs12
Hi,

I am using Abinit on both the computing clusters and my own desktop and they give vastly different results. An example of the output files from both systems using the same input is attached for your reference. Note that in hpc.out, the total energy jumps from one value to another very different value moving from one structural relaxation cycle to another, while in ubuntu.ac, the results are stable. I would like to understand the source of the problem, e.g. if I should change the way I configure Abinit in the computing cluster , or if the problem exists with the computing cluster itself like memory issue etc...

I compiled Abinit 8.10.1 on both systems and have tested them using tests/runtests.py without error. Here is the configuration file for the computing cluster that I used:

CC="mpicc"
CXX="mpic++"
FC="mpif90"

enable_mpi="yes"
enable_mpi_io="yes"
with_mpi_prefix=${OPENMPI_HOME}

with_dft_flavor="libxc"

with_trio_flavor="netcdf"
with_netcdf_incs="-I/central/software/netcdf-fortran/4.4.4/include -I/central/software/hdf5/1.10.1-ompi300/include"
with_netcdf_libs="-L/central/software/netcdf-fortran/4.4.4/lib -lnetcdff -L/central/software/hdf5/1.10.1-ompi300/lib -lhdf5_hl -lhdf5"

with_fft_flavor="fftw3"
with_fft_incs="-I/central/software/FFTW/3.3.7/include"
with_fft_libs="-L/central/software/FFTW/3.3.7/lib -lfftw3 -lfftw3f"

Thanks so much for your attention!

Re: Different calculation results for different systems

PostPosted: Mon Jan 28, 2019 12:33 pm
by ebousquet
Dear cgs12,
Could your question related to this post:
https://forum.abinit.org/viewtopic.php?f=3&t=3391&p=10358&hilit=avx#p10358
?
Eric

Re: Different calculation results for different systems

PostPosted: Sat Feb 02, 2019 12:06 am
by cgs12
The problem faced sounds very similar. I will try it out. Thanks!