[lammps-users] LAMMPS vs GROMACS

I was wondering whether anyone has benchmarked LAMMPS against GROMACS
for MD of molecular systems with Lennard-Jones and Coulombic

The reason i'm asking is that i've done some test runs using both
LAMMPS and GROMACS on Intel Core 2 Quad machine of a system of 2048
SPC/E water molecules (LJ site on oxygen, charged sites on oxygen and
2 hydrogens), periodically replicated in 3D, and I find that GROMACS
is roughly a factor of 2 faster. I get roughly the same result (i.e. a
factor of ~2 difference between LAMMPS and GROMACS) running the
calculation as a serial job and in parallel (4 processors) using
LAM/MPI. I've used pretty much the same run parameters and compiler
options for the two calculations.

Has anyone encountered such a large difference in the run speeds
between LAMMPS and GROMACS?

I would also appreciate it if anyone could suggest ways I could speed
up my LAMMPS calculation as I would prefer not to switch to GROMACS.

I've attached my LAMMPS input file and make file below (the starting
configuration was an fcc lattice of randomly oriented SPC/E water
molecules at the liquid density at 298 K and the run was carried out
long enough to reach equilibrium).


# LAMMPS input file

GROMACS has assembler versions of pairwise interactions and
also does special things with water. So a factor of 2x is not