Hi, I am currently using Lammps 2001 and I want to upgrade to the newest version of Lammps. Our research group have 64-bit processors for our new cluster however I am still running the old version of lammps with codes that I think are written for 32 bit processors. I know that I need MPI and FFTW for Lammps to work and I’v heard that FFTW for 64 bit is not yet available. I am hoping that you could help me exploit the 64-bit processors so I can run faster simulations. My typical simulations consists of 50K atoms running with short range and long range electrostatic potential via PPPM. And, it takes about three weeks to reach up to 1,000,000 MD Steps.
There is nothing I am aware of in the current LAMMPS that should
care about 32 vs 64-bit. We compile and run on 64-bit platforms
using the standard FFTW without any issues …
Steve
Jan-Michael,
The current version of LAMMPS should run roughly twice
as fast as LAMMPS 2001 for the problem you've
described (50k atoms, PPPM). The speedup is due mostly
to tabulation of the real-space portion of the
long-range electrostatics.
I'm surprised it takes 3 weeks to get a million
timesteps. If you run on around 50 fairly fast
processors with reasonable inter-processor
communication speeds, you should be able to get a
million timesteps in a couple of days.
Paul