Hey everyone,
I’m having a bit of trouble with GPU acceleration in LAMMPS 4Feb2025 and was hoping to get some input. Here’s what’s going on:
I’m using an RTX 4060 Ti, and when compiling LAMMPS, I enabled the GPU package (PKG_GPU=ON), configured the CUDA path correctly, and set the GPU architecture to sm_89 (which matches the card’s compute capability).
But when I run with this command:
mpirun -np 8 ~/Downloads/lammps-4Feb2025/build/lmp -in in.simulation -sf gpu -pk gpu 1
nvidia-smi shows 0% GPU utilization the entire time, and there’s no speedup compared to running without GPU acceleration.
I’ve checked the CMake logs to confirm PKG_GPU is enabled, the CUDA path is correct, and the executable (build/lmp) exists. Wondering if I messed up the command flags? Or do I need to specifically use GPU-enabled pair styles in the input script (like pair_style …/gpu)? Is there a way to check if LAMMPS is actually loading the GPU kernels?
Any help or ideas would be really appreciated—thanks!