I compiled LAMMPS with KOKKOS package (nvhpc23.11, cuda12). But when I ran the lammps script with the command “mpirun -np 1 lmp -k on g 1 -sf kk -pk kokkos neigh half -in base.in”, it failed to run. The error information is shown below:
"
LAMMPS (29 Aug 2024 - Update 1)
KOKKOS mode with Kokkos version 4.3.1 is enabled (src/KOKKOS/kokkos.cpp:72)
will use up to 1 GPU(s) per node
C++ memory allocation failed: (null)
MPI_ABORT was invoked on rank 0 in communicator MPI_COMM_WORLD
with errorcode 1.
NOTE: invoking MPI_ABORT causes Open MPI to kill all MPI processes.
You may or may not see output from other processes, depending on
exactly when Open MPI kills them.
"
The lammps version: 29 Aug 2024 - Update 1
GPU: A100
Without a simple way to reproduce what you are seeing, there is very little that we can do to help you. Chances are, that there is a problem with the size of a system that you want to simulate.
Here are a few things that you can try:
run LAMMPS with the inputs in the “bench” folder. Check the README file for instruction on how to grow the problems by using -var flags until it reaches the size of your system
check if you can run the same input without KOKKOS.
try using the latest feature release version 4 Feb 2025 of LAMMPS
Likely you are running out of memory. Does it run without KOKKOS enabled on a CPU? Like Axel said, we need a minimal working example that reproduces the issue to investigate further.