LAMMPS with GPUs and infinit

[wilkek]

I currently cannot make it run and I think it might be about this

[engine]
lmp = '/path/to/lmp'

what does not fully reflect the way I am running regular MD:
mpirun -np 1 /path/to/lmp -k on g 1 -sf kk -in input.lammps.

Can you give me advise here? Do I need to account for it in my slurm script maybe?

[dz24]

I have run lammps with GPU before, but only with 1 worker so far. Here is how I worded the command:

[engine]
lmp = "mpirun lmp -sf gpu"

I assume you can put the whole thing in, except for maybe -in input.lammps as infretis likely adds this itself.

[wilkek]

Okay good. Do you run the worker with a separate srun as one does with ase? And is there a reason that the lammps interface is not used more? Is it the file size?

[wilkek]

I could make it work on multiple workers. The only issue is that the energy.txt only contains nan instead of values. Is this a known issue?

[dz24]

Okay good. Do you run the worker with a separate srun as one does with ase?

Depends on the HPC. in the HPC I had there is no srun, only mpirun.

And is there a reason that the lammps interface is not used more? Is it the file size?

I think partly file size, but also more people may currently be working on biological systems

I could make it work on multiple workers. The only issue is that the energy.txt only contains nan instead of values. Is this a known issue?

Did you use latest the infretis version? I think it is fixed by a merged PR

infretis/infretis#180

[wilkek]

Did you use latest the infretis version? I think it is fixed by a merged PR

infretis/infretis#180

Yes I do, and also the rest of the columns shows nan

# Cycle: 8, status: ACC, move: ('wf', 9000, 3, 11)
Time      Potential        Kinetic          Total    Temperature
 0            nan            nan            nan            nan

[lukasbaldauf]

How does your lammps input script look? It might be related to the lammps sections defining the energy output.

[wilkek]

Yes indeed it was. In the infretis/examples/lammps input, there was this line missing:

thermo_modify flush yes

Now the last issue is, that the generated temperature in my nVE is too low. I have set it to 300K in the infretis.toml, but the generated velocities do not reflect that.

[wilkek]

Oh I found out - I am using units metal, which has velocities in Ang/ps but infRETIS uses units real with Ang/fs in the velocity generation

[lukasbaldauf]

Good, you found out! The engine hasn't been tested a lot, so there are a lot of these tiny catches atm unfortunately. Usually, I start with the lammps.input file from the examples/lammps/H2 system, and then add the lines I need.