What is the "pair" operation? Does it only do it once?

Hi, All.

I am new lammps user.
I have some questions about the GPU packages.

  1. I have always seen a lot of time spent on “Pair” operation. What does this operation do?

  2. For GPU, it seems each step it only transfers a little data into GPU. But actually, the GPU memory consumption is much larger. What are the data sent to GPU in each timestep? And what data is transferred back to CPU and what operation is done?

Best,
Dong

https://docs.lammps.org/Run_output.html
https://docs.lammps.org/pair_style.html

https://docs.lammps.org/Speed_gpu.html

@Article{Brown11,
 author = {W. M. Brown and P. Wang and S. J. Plimpton and A. N. Tharrington},
 title = {Implementing Molecular Dynamics on Hybrid High Performance Computers---Short Range Forces},
 journal = {Comput.\ Phys.\ Commun.},
 year =    2011,
 volume =  182,
 pages =   {898--911},
 doi =     {10.1016/j.cpc.2010.12.021}
}

Thanks for the reply.

" https://docs.lammps.org/Run_output.html "
If I use the GPU package, will this computation(Pair) happen on GPU?

I am curious about the workflow of GPU packages.
In the output, they give two breakdown time.

Here, we got MPI time and device info.
Does MPI time include the device time?

The command is
mpiexec -np 20 …/build/lmp -sf gpu -pk gpu 1 neigh yes newton off -in in.example1

Thanks in advance.

If you want to know whether GPUs are used, you can just run without and compare the output.

I have nothing else to add to the information provided by the documents I pointed out to you. If you need more details, you need to study the source code.