With version 6.5 from github,building on trestles to use the oldest environment for xHOST:
module load intel/19.0.5 impi/19.0.5 mkl/19.0.5 $ ./install/configure \ MPIF90=mpiifort F90=ifort F77=ifort FC=ifort CC=icc \ SCALAPACK_LIBS="-lmkl_scalapack_lp64 -lmkl_blacs_intelmpi_lp64" \ LAPACK_LIBS="-lmkl_lapack95_lp64 -lmkl_blas95_lp64" \ BLAS_LIBS="-lmkl_intel_lp64 -lmkl_intel_thread -lmkl_core" \ FFT_LIBS="-L${MKL_ROOT}/interfaces/fftw3xf -lfftw3xf_intel" \ FCFLAGS="-O3 -xHOST -axsse4.2,AVX,CORE-AVX512 -assume byterecl" \ CCFLAGS="-O3 -xHOST -axsse4.2,AVX,CORE-AVX512 -assume bytrerecl \ -D__INTEL -D__GNUC__ -D__FFTW3 -D__MPI -D__PARA -D__SCALAPACK" \ --with-hdf5=/share/apps/hdf5/1.10.5/intel/impi \ --with-scalapack=intel --enable-openmp --enable-parallel \ --prefix=/share/apps/espresso/espresso-6.5-intel-impi-mkl make pw make all
Using the AUSURF112 benchmark, running on single nodes of trestles and razor, and last a comparison run with qe 6.1:
OMP_NUM_THREADS=1 mpirun -np 32 /scrfs/apps/espresso/espresso-6.5-intel-mkl-impi/bin/pw.x -npools 1 <ausurf.in OMP_NUM_THREADS=1 mpirun -np 32 /scrfs/apps/espresso/espresso-6.5-intel-mkl-impi/bin/pw.x -npools 2 <ausurf.in OMP_NUM_THREADS=1 mpirun -np 16 /scrfs/apps/espresso/espresso-6.5-intel-mkl-impi/bin/pw.x -npools 1 <ausurf.in OMP_NUM_THREADS=1 mpirun -np 16 /scrfs/apps/espresso/espresso-6.5-intel-mkl-impi/bin/pw.x -npools 2 <ausurf.in OMP_NUM_THREADS=2 mpirun -np 8 /scrfs/apps/espresso/espresso-6.5-intel-mkl-impi/bin/pw.x -npools 1 <ausurf.in mpirun -np 16 /share/apps/espresso/qe-6.1-intel-mkl-impi/bin/pw.x -npools 1 <ausurf.in
with results showing 6.5 much improved over 6.1.
Run | System | OMP | MPI | npools | walltime |
---|---|---|---|---|---|
1 | trestles | 1 | 32 | 1 | 867 |
2 | trestles | 1 | 32 | 2 | 837 |
3 | razor | 1 | 16 | 1 | 690 |
4 | razor | 1 | 16 | 2 | did not run |
5 | razor | 2 | 8 | 1 | 1073 |
6 | razor qe 6.1 | 1 | 16 | 1 | 1612 |