Thanks for replying Alex. I've used this input file on a supercomputer
before and it works just fine. I also tried to run it on my CPU but it takes
there was a bug in lammps that is fixed by the very last patches,
but not yet in the git or svn repo. those will have the changes
yes, you will see a speedup, but there will be *more* of
a speedup, when using less MPI tasks. hyperthreading
helps very little on the CPU already. i see about 10%
speedup when going from 8 MPI tasks to 16 MPI tasks
on my desktop.
that being said, running a test system with 32,000 eam
atoms takes a little over a second on my machine and
running the same on the GPU takes 0.88 seconds with
just one GTX 480 and one MPI task, but is not much
faster with 0.86 seconds and two MPI tasks. using
more MPI tasks, slows it down again.