[gmx-users] Problems with mdrun (version 4.6.5) and CPU-type: Xeon E5472

[Thomas Schlesier]

Thank you for the suggestion. That solved the problem.

If I understand it correctly GMX_DISTRIBUTABLE_BUILD=ON only disables 
the support for 'rdtscp', which is used for determining the wall-time. 
Does this effect the determination of the wall-time? Meaning when one 
uses GMX_DISTRIBUTABLE_BUILD=ON, will then an other instruction be used 
for the determination and is that one also accurate?

Greetings
Thomas


Am 24.01.2014 19:54, schrieb 
gromacs.org_gmx-users-request at maillist.sys.kth.se:
> Please try setting GMX_DISTRIBUTABLE_BUILD=ON for CMake and see whether
> that fixes the problem. On Fri, Jan 24, 2014 at 6:18 PM, Thomas
> Schlesier <schlesi at uni-mainz.de>wrote:
>> >Hi all,
>> >I am having problems with GROMACS 4.6.5 and one type of CPU on our cluster
>> >(other CPU-types make no problems). GROMACS was always compiled on the
>> >master, which uses a different CPU-type (see below at the beginning of the
>> >'md.log'-file).
>> >
>> >For the CPU type (partial output from 'cat /proc/cpuinfo'):
>> >vendor_id       : GenuineIntel
>> >cpu family      : 6
>> >model           : 23
>> >model name      : Intel(R) Xeon(R) CPU           E5472  @ 3.00GHz
>> >stepping        : 6
>> >cpu MHz         : 3000.186
>> >cache size      : 6144 KB
>> >
>> >'grompp' 'g_rdf' and 'pdb2gmx' all work fine, only 'mdrun' gives problems
>> >(so I think only 'mdrun' is affected).
>> >
>> >Error message is:
>> >Illegal instruction
>> >and i get a 'core dump'. (Didn't looked into the dumped file, since i only
>> >know i can do this with 'gdb' but have absolute no idea how 'gdb' works and
>> >for what i should look).
>> >
>> >Last lines of the 'md.log' file before the crash (energy minimization of a
>> >small water box with cut-offs):
>> >
>> >Removing pbc first time
>> >
>> >++++ PLEASE READ AND CITE THE FOLLOWING REFERENCE ++++
>> >S. Miyamoto and P. A. Kollman
>> >SETTLE: An Analytical Version of the SHAKE and RATTLE Algorithms for Rigid
>> >Water Models
>> >J. Comp. Chem. 13 (1992) pp. 952-962
>> >-------- -------- --- Thank You --- -------- --------
>> >
>> >So, i get the impression, that the preparation for the simulation is
>> >success full, and the crash happens when the actual simulation starts.
>> >
>> >The same simulation works with gromacs 4.0.7 and 4.5.5 (on the same
>> >machine). So I thought SSE4.1 or the internal MPI-stuff from gromacs 4.6.5
>> >produces the problem.
>> >So i compiled gromacs 4.6.5 with
>> >-DGMX_CPU_ACCELERATION=SSE2 (or) None
>> >and without any MPI support (and then running only a serial calculation).
>> >Also tried for FFT the internal 'fftpack' (thought maybe the FFT part
>> >might be the problem - for this i also ran the simulation without PME and
>> >only plain cut-offs).
>> >
>> >But the error still occurs in every case:(
>> >
>> >I tried the compilation with two different compilers:
>> >Intel:
>> >icc (ICC) 12.1.2 20111128
>> >Copyright (C) 1985-2011 Intel Corporation.  All rights reserved.
>> >and GNU:
>> >gcc (GCC) 4.6.4 20120511 (prerelease)
>> >Copyright (C) 2011 Free Software Foundation, Inc.
>> >but i makes no difference.
>> >
>> >
>> >To give some more information:
>> >
>> >Beginning of the 'md.log'-file:
>> >
>> >Log file opened on Fri Jan 24 16:40:34 2014
>> >Host: tc-3-12.local  pid: 6755  nodeid: 0  nnodes:  1
>> >Gromacs version:    VERSION 4.6.5
>> >Precision:          single
>> >Memory model:       64 bit
>> >MPI library:        none
>> >OpenMP support:     disabled
>> >GPU support:        disabled
>> >invsqrt routine:    gmx_software_invsqrt(x)
>> >CPU acceleration:   SSE4.1
>> >FFT library:        fftpack (built-in)
>> >Large file support: enabled
>> >RDTSCP usage:       enabled
>> >Built on:           Fr 24. Jan 16:26:42 CET 2014
>> >Built by:schlesi at dirac.chemie.uni-mainz.de  [CMAKE]
>> >Build OS/arch:      Linux 2.6.18-238.19.1.el5 x86_64
>> >Build CPU vendor:   GenuineIntel
>> >Build CPU brand:    Intel(R) Xeon(R) CPU           X5670  @ 2.93GHz
>> >Build CPU family:   6   Model: 44   Stepping: 1
>> >Build CPU features: apic clfsh cmov cx8 cx16 htt lahf_lm mmx msr
>> >nonstop_tsc pcid pdcm pdpe1gb popcnt pse rdtscp sse2 sse3 sse4.1 sse4.2
>> >ssse3
>> >C compiler:         /share/apps/gcc/gcc-4.6.4_20120511/bin/gcc GNU gcc
>> >(GCC) 4.6.4 20120511 (prerelease)
>> >C compiler flags:   -msse4.1    -Wextra -Wno-missing-field-initializers
>> >-Wno-sign-compare -Wall -Wno-unused -Wunused-value -Wno-unknown-pragmas
>> >-fomit-frame-pointer -funroll-all-loops -fexcess-precision=fast  -O3
>> >-DNDEBUG
>> >
>> >cmake-command for compilation (used for the last build):
>> >(here i'm using line breaks to make it easier to read):
>> >
>> >cmake
>> >-DCMAKE_C_COMPILER=/share/apps/gcc/gcc-4.6.4_20120511/bin/gcc
>> >-DCMAKE_CXX_COMPILER=/share/apps/gcc/gcc-4.6.4_20120511/bin/g++
>> >-DGMX_BUILD_OWN_FFTW=OFF
>> >-DCMAKE_INSTALL_PREFIX=/scrcluster/schlesi/bug465 -DBUILD_SHARED_LIBS=OFF
>> >-DGMX_PREFER_STATIC_LIBS=ON
>> >-DGMX_CPU_ACCELERATION=SSE4.1
>> >-DGMX_MPI_IN_PLACE=OFF
>> >-DGMX_OPENMP=OFF
>> >-DGMX_THREAD_MPI=OFF
>> >-DGMX_MPI=OFF
>> >-DGMX_FFT_LIBRARY=fftpack
>> >..
>> >(Sidenote: the 'GMX_MPI_IN_PLACE' isn't used - cmake tells me this).
>> >
>> >Till now i havn't tried to compile GROMACS on any of the problematic
>> >machines, but i can try this on monday.
>> >
>> >
>> >------------------------------------------------------------------------
>> >
>> >Has anyone any idea (for solving, or to better trace the error)? Since i'm
>> >at the end with my knowledge (actually i'm quite happy that i came so far,
>> >since about this 'hardcore' computer stuff my knowledge is very limited).
>> >If someone needs more informations (or the core dumped file), just tell me.
>> >
>> >Greetings (and wish you a nice weekend)
>> >Thomas