Insufficient memory on accelerator on GPU

LAMMPS LAMMPS Installation
,
21

How did you build with CUDA? Can you share the cmake command, or the steps if you build with traditional makefile build?

22

Absolutely!

I have used [email protected] and [email protected]. We have RTX 3090 Blower GPUs if that is relevant.

cmake ../cmake -D LAMMPS_MACHINE=gpu -D PKG_OPENMP=yes -D PKG_PLUMED=yes -D DOWNLOAD_PLUMED=yes -D PKG_MPI=yes -DPKG_OMP=yes -D PKG_EXTRA-FIX=yes -D PKG_MOLECULE=yes -D PKG_H5MD=yes -D PKG_EXTRA-PAIR=yes -D PKG_MANYBODY=yes -D PKG_DPD-BASIC=yes -D PKG_RIGID=yes -D PKG_KSPACE=yes -D PKG_GPU=yes -D GPU_API=cuda -D GPU_PREC=mixed -D GPU_ARCH=sm_86

cmake ../cmake
cmake --build . -j

Thank you!

23

cuInit is from the CUDA driver API, do you have the path to libcuda.so in LD_LIBRARY_PATH when you run LAMMPS?