# lammps最好紀錄 ## lj ### cpu intel package ```shell= cmake/3.14.5 mkl/2019u4 intel/2019u4 openmpi/4.0.1 ``` ``` cmake -D BUILD_OMP=yes -D BUILD_MPI=yes -D PKG_MOLECULE=yes -D PKG_RIGID=yes -D PKG_KSPACE=yes -D PKG_USER-OMP=yes -D PKG_USER-INTEL=yes -D INTEL_ARCH=cpu -D INTEL_LRT_MODE=c++11 -D FFT=MKL -D FFT_SINGLE=no ../cmake ``` 4n-40ppn-2omp ``` mpirun --bind-to none --map-by node -mca pml ucx --mca btl ^vader,tcp,self -mca coll_hcoll_enable 1 -x HCOLL_MAIN_IB=mlx5_0:1 -x UCX_NET_DEVICES=mlx5_0:1 -x UCX_IB_PKEY=0x0003 $LMP < $INPUT -sf intel -pk intel 0 mode double ghost yes ``` ~~->後來跑出來的似乎都達不到最快~~ hyperthreading目前最快 ``` #SBATCH --ntasks-per-node=80 mpirun --bind-to none --map-by node -mca pml ucx --mca btl ^vader,tcp,self -x UCX_NET_DEVICES=mlx5_0:1 -x UCX_IB_PKEY=0x0003 $LMP < $INPUT -sf intel -pk intel 0 mode double ghost yes ``` ### gpu ## rhodo ### cpu opt package ```shell= cmake/3.14.5 gcc/8.3.0 openmpi/4.0.1 fftw/3.3.8 ``` ``` cmake -D BUILD_OMP=yes -D BUILD_MPI=yes -D PKG_MOLECULE=yes -D PKG_RIGID=yes -D PKG_KSPACE=yes -D PKG_OPT=yes ../cmake ``` 4n-80ppn ``` mpirun --bind-to none $LMP < $INPUT -sf opt ``` ###### tags: `lammps`