Dear LAMMPS developers,
is there any of you how has worked on an "optimized" version of LAMMPS for IBM BlueGene/Q machine at CINECA(italy)? Unlikily, I was asked to install a centralized version of LAMMPS on this kind of architecture and, frankly, I don't see further "magic" I can cast but "adapting" makefiles. Any useful hint?
dear luca,
Dear LAMMPS developers,
is there any of you how has worked on an "optimized" version of LAMMPS
for IBM BlueGene/Q machine at CINECA(italy)? Unlikily, I was asked to
install a centralized version of LAMMPS on this kind of architecture
and, frankly, I don't see further "magic" I can cast but "adapting"
makefiles. Any useful hint?
you should probably compare notes with adrian langer from argonne,
who has been benchmarking and trying to improve LAMMPS for
bluegene type architectures.
you should make sure that the user-omp package is installed
and provide instructions how to correctly run LAMMPS in
hybrid MPI+OpenMP mode.
for people running with long-range electrostatics, may also
look into using the verlet/split integrator, which runs kspace
on separate MPI tasks. unfortunately, i have not found the time
yet to make this properly OpenMP compatible.
ciao,
axel.