Cuda error when using package DPD-BASIC

Dear all, it’s my first time writing on this forum so I apologize in advance if I am not opening this topic in the right way.
As already stated in the object, I have built lammps with make (version 2 Aug 2023) for the gpu package.
Running the ~/bench/POTENTIALS/in.dpd with the command
$ mpirun -np 4 lmp_mpi -sf gpu -pk gpu 1 -i in.dpd
I get the error:
“Cuda driver error 700 in call at file ‘geryon/nvd_timer.h’ in line 76”
I’ve seen a similar issue in the forum (“Problem with Cuda”) where @akohlmey suggested to try and check also the “in.rhodo” benchmark file. So I did it and it worked.
That’s why I’m writing to ask you if there is anything else I can try to test my lammps build or if there is a known issue with this package that I am not aware of.
I link all necessary files to see my OS, architecture, drivers, LAMMPS package installed but also the output of the benchmarks cited above.

PS: the last version of LAMMPS that I was using in the same machine with the same build (except for the cuda version which was 11-7 back then) was the one from 2 Jun 2022 and everything used to work just fine.

I hope someone can help me with this issue, thanks in advance

Giorgio Turtu’
dpd_nogpu.out (2.3 KB)
dpd_withgpu.out (1.4 KB)
dpd_withgpu_error.out (976 Bytes)
lscpu.out (2.7 KB)
Makefile.mpi (2.0 KB)
nvidia_smi.out (2.2 KB)
package_installed.out (218 Bytes)
rhodo_nogpu.out (4.6 KB)
rhodo_withgpu.out (6.7 KB)

This is a known bug that was fixed in the development version and in the maintenance branch, but no release has been made that fixes this bug.

Thus you would have to either download the maintenance branch from GitHub, or the development snapshot

or just apply the fix from this commit bugfix for pair style dpd/gpu from Trung · lammps/lammps@2f415c8 · GitHub
and recompile.

1 Like

Thanks @akohlmey for the reply, I have modified the lines in the “lal_dpd.cu” file and it worked.

Giorgio Turtu’