[gmx-users] CygWin and Gromacs 4.5.5

[Mirco Wahab]

On 13.11.2011 02:10, Mirco Wahab wrote:
> I was able to build Gromacs 4.5.5 (32 bit) on Cygwin (gcc 4.5.3)
> w/o threads

For a better understanding of the usability of
cygwin-compiled Gromacs under windows I performed
some more tests on my "desktop box".

First, I tried to install recent versions of MPICH2 or
OpenMPI in cygwin for creating a mdrun_mpi binary. Both
MPI implementations won't compile under cygwin alone
anymore (will require MSVC tools). Therefore, I tested
the single-threaded mdrun version under cygwin. The result
is compared with Gromacs compiled w/other toolchains on
the same desktop machine. To compare architectures,
runtime results on a cluster node (i7) are provided
for the same simulated system.

A. Simulated system
------------------------------
50nm Box w/coarse grained solvent (MARTINI W) and
10K coarse grained lipids (MARTINI DPPC), total
number of particles ~ 1 Mio. No Ewald or PME, simple
electrostatics w/shifted coulomb and rlist=1.3nm,
large dt=0.04, v-rescale and isotropic Berendsen
pcoupl. Large, but simple system. Gromacs 4.5.5
has been compiled from source "as good as possible".
The GFlops values have been copied from mdrun output
ending after 2000 steps (or 200 steps for cygwin-mdrun).

B. Box at Home: PhenomII/X6 3.4GHz, 4GB DDR3, Windows 7 Pro/x64
------------------------------------------------------------
No  GFlops  System
------------------------------------------------------------
1 | 11.3 | Win64 native (MSVC 2010 SP1 + SDK 7.1 SP1), 6 Threads, 3x2x1
2 | 11.1 | SuSE 12.1rc (GCC 4.6) as VMWare Guest OS, 6 Threads, 3x2x1
3 | 0.49 | CygWin 1.7.9-1/Setup 2.738 (GCC 4.5.3), Single Thread, no DD

For the cygwin run, the mdrun.exe process has been allocated
to one cpu core and its priority was set to 'high'. These
"optimizations" didn't change the outcome at all. Maybe
on a Win32 box (instead of Win64), the situation is different?

C. For comparison: i7/2600K @4.2GHz, 4GB DDR3, SuSE 11.3/x64
------------------------------------------------------------
No  GFlops  System
------------------------------------------------------------
  1 | 14.0 | GCC  4.5.1 (4 Threads, 4 x 1 x 1)
  2 | 14.3 | ICC 12.0.4 (4 Threads, 4 x 1 x 1)
  3 | 19.6 | GCC  4.5.1 (8 Threads, 4 x 2 x 1)
  4 | 19.9 | ICC 12.0.4 (8 Threads, 4 x 2 x 1)


Regards,

M.