Hello lammps users:
I was trying to compile lammps-1Feb-14 using package USER-OMP, OPT and GRANULAR and it it failing.
I know this is a known errors.
I would like to now if this has been already solved or not.
Also I would like to know which option is better in an single multicore CPU environment ( Intel i7-4700MQ with 8 cores) either using MPI or OpenMP. Some insight on this??
Hello lammps users:
I was trying to compile lammps-1Feb-14 using package USER-OMP, OPT and
GRANULAR and it it failing.
I know this is a known errors.
I would like to now if this has been already solved or not.
yes. it has been solved.
Also I would like to know which option is better in an single multicore
CPU environment ( Intel i7-4700MQ with 8 cores) either using MPI or
OpenMP. Some insight on this??
there is not a simple answer to this. which is better depends very
much on your pair style and system setup. for dense, homogeneous
systems, MPI is often the most effective parallelization, since LAMMPS
was designed to handle this case well. however, OpenMP parallelization
is less affected by load imbalance and the /omp pair styles have a lot
of serial optimizations. so in practice using OpenMP with 2 or 3 (or
1!) threads on top of MPI is often faster.
axel.
Thanks for your fast reply.
Hello lammps users:
I was trying to compile lammps-1Feb-14 using package USER-OMP, OPT and
GRANULAR and it it failing.
I know this is a known errors.
I would like to now if this has been already solved or not.yes. it has been solved.
Also I would like to know which option is better in an single multicore
CPU environment ( Intel i7-4700MQ with 8 cores) either using MPI or
OpenMP. Some insight on this??there is not a simple answer to this. which is better depends very
much on your pair style and system setup.
I am using gran/hertz/history pair style with opt package that claims to run faster because the implementation uses the c++ template mechanism. ( i don't know if this package indeed provide some speed up on my particular pair style). An i am simulation a very dense system.
there is no OPT version of gran/hertz/history
axel.