Dear all,
I have LAMMPS (29 Aug 2024 - Update 1) version installed with CUDA 12.2. I am testing the wetting and evaporation process of coarse-grained water molecules on the surface of alloys. Among them, SW potential is used for coarse-grained water molecules, EAM/alloy potential is used between alloys, and simple LJ potential is used between metal atoms and water molecules. Therefore, I used hybrid/overlay to apply different types of forces to atoms.
pair_style hybrid/overlay sw eam/alloy lj/cut 12
The attachment contains my input file and the potential function file I used.
FeNiCr.t (570.1 KB)
in.random (3.6 KB)
inconel600.data (3.4 MB)
mW.sw (509 Bytes)
Due to the error message “ERROR: Must not use GPU neighbor lists with hybrid pair style” when using GPU directly. According to Jasmine 969’s description on the forum("ERROR: Must not use GPU neighbor lists with hybrid pair style" when running lammps on GPU from Python), this issue can be resolved by adding ‘neigh no’. When I use input commands such as “mpirun -np 8 lmp_mpi -sf gpu -pk gpu 1 neigh no -in in.random”, similar problems will not occur again.
But after completing the miniaturization, when officially entering the simulation, the following error will occur:
WARNING: Communication cutoff adjusted to 15 (…/comm.cpp:739)
Cuda driver error 700 in call at file ‘geryon/nvd_memory.h’ in line 502.
Abort(-1) on node 4 (rank 4 in comm 0): application called MPI_Abort(MPI_COMM_WORLD, -1) - process 4
Due to the unacceptable slow computing speed when using only the CPU, and the inability to scale to larger systems. Have any scholars encountered similar problems and successfully solved them?
Many thanks in advance for any guidance on this issue