Erik Lindahl <lindahl [at] gromacs.org>
Department of Structural Biology
Stockholm Bioinformatics Centre
David van der Spoel <spoel [at] gromacs.org>
Department of Biochemistry
Box 576, 751 23 Uppsala
Benchmark Program General Category
Chemistry / Molecular Dynamics
435.gromacs is derived from GROMACS, a versatile package that performs molecular dynamics,
i.e. simulation of the Newtonian equations of motion for systems with
hundreds to millions of particles.
Although it is primarily designed for
biochemical molecules such as proteins and lipids that have many
complicated bonded interactions, GROMACS is also extremely
fast at calculating the nonbonded interactions (that usually
dominate the simulation cost). Therefore, many groups are also using it for
research on non-biological systems, such as polymers.
The benchmark version performs a simulation of the protein Lysozyme in
a solution of water and ions. The structure of a protein is
normally determined by experimental techniques such as X-ray
crystallography of NMR spectroscopy. By simulating the atomic
motions of these structures, one can gain significant understanding
of protein dynamics and function, and, in some cases, it might even
be possible to predict the structure of new proteins.
A dodecahedron-shaped box is used to reduce the amount of solvent
water, but there are still 23179 atoms in the system. The
simulation time is completely dominated by the inner loops where
the nonbonded Lennard-Jones and Coulomb interactions are calculated
between atoms that are closer than 1.4 nm in space.
There are three different run input files with identical setup,
but a different number of steps taken.
- test 1500 steps
- train 3000 steps
- ref 6000 steps
Once the code is compiled the SPEC harness invokes the
benchmark with a command such as:
gromacs -nice 0 -deffnm gromacs -silent
The -deffnm option uses the name provided for the base of all
input and output files (in this case “gromacs”): gromacs.tpr is the input file and
gromacs.out is the output file.
The nice flag has no effect
on Windows, but on Unix the program defaults to nice 19 if it is
not given. The -silent option turns off producing a log file.
For validation a couple of energy terms are written in clear
text to the file gromacs.out. These are the average potential energy,
the average system temperature, and the number of floating point
Molecular dynamics is by definition a chaotic process.
Normally, quite agressive compiler optimizations are used to compile the code,
hence slight variations are
quite normal. The gromacs.out results shouldn’t differ by more than 1.25%
from the reference values.
C and Fortran. The only Fortran code is the inner loops
(innerf.f). The inner loops typically account for more than 95% of
the runtime, so all the computationally intensive parts use Fortran.
Known portability issues
If a Fortran compiler does not append an underscore to
external names, the flag
can be used to
change the name mangling macro in config.h.
SPEC has chosen some settings in config.h to select
whether one’s C compiler supports strdup(),
strcasecmp(), popen()/pclose(), or if it lacks some ANSI C keywords
like ‘const’. None of these will affect performance, but it makes
the benchmark compatible with standard Gromacs input/output
files. If a compiler should come along that is incompatible
with SPEC’s settings in config.h, SPEC should be notified (see
Portability issues fixed subsequent to the release of SPEC CPU2006 V1.0
For CPU2006 V1.1:
Internal system calls which set the OS scheduling priority were disabled. The effect of this change is minimal,
because when it is run from the SPEC harness, the GROMACS line option ‘-nice 0’ is included, which also disables these
system calls. For V1.0, a user studying 435.gromacs outside the runspec harness might have encountered inconsistent
measurements if ‘-nice 0’ was not used.
Replaced a GPL version of the C99 math function “erfc” with a public
Added a new portability flag ‘-DSPEC_CPU_HAVE_ERF‘ which can be set to cause use of a
system’s “erfc” library function instead of the version included with the benchmark. This flag is the default on some
You can find more information and documentation at www.gromacs.org
Last updated: $Date: 2008-04-12 08:31:17 -0400 (Sat, 12 Apr 2008) $
Read more here: Source link