Skip to content

Latest commit

 

History

History
33 lines (32 loc) · 2.25 KB

install_gromacs.md

File metadata and controls

33 lines (32 loc) · 2.25 KB
Raw

CPU installation using intel compiler

  • Reference: https://software.intel.com/en-us/articles/recipe-building-and-running-gromacs-on-intel-processors
  • Download the latest gromacs (2019.2 as of April 2019
  • Set up intel compiler, mpi, mkl
  • Unpack gromacs source zip and go to main folder
  • mkdir build
  • cd build
  • FLAGS="-xCORE-AVX512 -g -static-intel"; CFLAGS=$FLAGS CXXFLAGS=$FLAGS CMAKE_C_COMPILER=mpiicc
    CMAKE_CXX_COMPILER=mpiicpc CC=mpiicc CXX=mpiicpc cmake .. -DBUILD_SHARED_LIBS=OFF -DGMX_FFT_LIBRARY=mkl
    -DCMAKE_INSTALL_PREFIX=/myproject/MD -DGMX_MPI=ON -DGMX_OPENMP=ON -DGMX_CYCLE_SUBCOUNTERS=ON -DGMX_GPU=OFF
    -DGMX_BUILD_HELP=OFF -DGMX_HWLOC=OFF -DGMX_SIMD=AVX_512 -DGMX_OPENMP_MAX_THREADS=32
  • make -j 32
  • Will produce build\bin\gmx_mpi. Run './build_atom/bin/gmx_mpi -version`
  • For centOS6.X, -std=c++11 is not met, and will not work. Install gcc >=5, 6,7, 8, and load them with intel compiler, so intel compiler can use those newer gcc library
  • Sample run command
    • Download Test Case A of Gromacs from http://www.prace-ri.eu/ueabs/
      • export OMP_NUM_THREADS=1
      • mpirun -np 8 /myproject/MD/gromacs-2019.2/build/bin/gmx_mpi mdrun -s ion_channel.tpr -maxh 0.5 -noconfout -nsteps 10000 -g logfile

With CUDA

  • Ref: http://manual.gromacs.org/documentation/current/install-guide/index.html
  • https://www.nvidia.com/en-us/data-center/gpu-accelerated-applications/gromacs/
  • nvcc is not used
  • mkdir gromacs-VERSION-build
  • cd gromacs-VERSION-build
  • CC=mpicc CXX=mpicxx cmake .. -DGMX_OPENMP=ON -DGMX_GPU=ON -DGPU_DEPLOYMENT_KIT_ROOT_DIR=/usr/nic/libs/cuda/10.0 -DGMX_MPI=ON -DGMX_BUILD_OWN_FFTW=ON -DGMX_PREFER_STATIC_LIBS=ON -DCMAKE_BUILD_TYPE=Release -DGMX_BUILD_UNITTESTS=ON -DCMAKE_INSTALL_PREFIX=/work/jeonb/MD -DGMX_HWLOC=OFF
    • If HWLOC=auto, openmpi may fail to compile
  • make -j 32
  • Running Gromacs
    • CPU only: OMP_NUM_THREADS=1 mpirun -np 40 gromacs-2019.2/build_cuda/bin/gmx_mpi mdrun -s ion_channel.tpr -maxh 0.5 -noconfout -nsteps 1000 -nb cpu
    • CPU+GPU: OMP_NUM_THREADS=2 mpirun -np 20 gromacs-2019.2/build_cuda/bin/gmx_mpi mdrun -s ion_channel.tpr -maxh 0.5 -noconfout -nsteps 1000 -nb gpu
      • Tuning/optimization is necessary