Hi, since I observe what I believe is the same problem, I'd like to make this thread alive again.
I tested installation from Fedora 25 repositories (cp2k version 3.0 svn:16458) on 2 notebooks, one with kernel 4.8.16 (8GB ram, 2-core CPU), another with kernel 4.9.3 (20GB ram, 4-core cpu). On the one with kernel 4.9.3 I compiled the trunk version 17688 as well (requirements installed from the repositories, see attached arch/Linux-x86-64-gfortran.popt makefile). Both openmpi and mpich versions result in the same behavior.
Some calculations run seemingly OK, but when I tried cp2k/tests/QS/regtest-hybrid-4 (or hybrid-2, I didn't try more) and try to run tests there I get:
#0 0x7fcddd807e3a
#1 0x7fcddd80702d
#2 0x7fcddccfb98f
#3 0x0
output file ends with:
Step Update method Time Convergence Total energy Change
------------------------------------------------------------------------------
Electronic density on regular grids: -7.9999964703 0.0000035297
Core density on regular grids: 7.9999969792 -0.0000030208
Total charge density on r-space grids: 0.0000005089
Total charge density g-space grids: 0.0000005089
HFX_MEM_INFO| Est. max. program size before HFX [MiB]: 80
Running the same test on our local cluster with trunk version 17176 goes fine. I can try to compile and run newer version on our cluster if that helps.
Any help is highly appreciated. Thank you
stano.
On Tuesday, July 5, 2016 at 9:36:35 AM UTC+2, Samuel Andermatt wrote: