I would like to ask for some clarification regarding the performance of my current LAMMPS setup. I am using LAMMPS 2024, as shown in the attached screenshots.
Could you please let me know if the configuration appears correct and whether the performance is expected for this version?
Additionally, I would like to know if there are recommended methods to further speed up simulations (e.g., using GPU/OMP/MPI optimizations or specific build options).
If upgrading to a newer or optimized build is advisable, could you please guide me on the best way to update, or point me to a recommended installation or build script?
Thank you very much for your time, assistance, and valuable work on LAMMPS.
Any advice or suggestions would be greatly appreciated.
Unlike many other MD softwares, in LAMMPS different force fields, integrators, etc. are vastly different in terms of performance or GPU support (because LAMMPS contains so many features and they are contributed by many different people). So it’s basically impossible to tell how much performance gain you can get without the full input deck. I recommend read the [accelerate performance] part of the documentation and understand how acceleration packages in LAMMPS work.
With that saying, judging from your limited system size (only 32000 atoms) I would say that GPU acceleration is likely not very useful. I think your best bet is to see if /intel works for you (again it depends on the whole simulation setup). Read the documentation for things to take care when running with the intel package.
If you have so much appreciation for the work of the LAMMPS developers, why don’t you demonstrate it by following the advice we are giving here in the forum (See the forum guidelines and suggestions post) and in the LAMMPS manual.
Screen shots are a bad idea.
Neither can be inferred from the information available in the image.
There is a discussion in the LAMMPS manual about how to measure and compare performance.
Please also note that “optimal performance” is no absolute item independent of boundary conditions but rather an optimization problem with very specific boundary condition (hardware, compiler, parallelization, input settings, etc.).
There are discussions on the topic in the LAMMPS manual. As explained in the guidelines, the LAMMPS developers very much appreciate if people would consult the manual first before posting questions. We write this stuff down, so we don’t have to explain the same (basic) things over and over again and so that for a future discussion there is a common ground.
Again, all the information you should need is in the LAMMPS manual, and what is best strongly depends on what your needs are. We put significant effort into documenting what LAMMPS can do and what we do to extend and improve it. It is all available. Do not expect that we can provide you with a digested version for your specific needs. We don’t have the time and even if there was time, we don’t know anything about your needs or available resources and skills. So how should anybody be able to provide anything specific.
Here you go. There are plenty of advice and suggestions in my post and even more in the other documents I am referring to. Please enjoy and educate yourself so we won’t have to complain in the future.
