2. Benchmarks #
3. in.intel.lj - Atomic fluid (LJ Benchmark)
4. in.intel.rhodo - Protein (Rhodopsin Benchmark)
5. in.intel.lc - Liquid Crystal w/ Gay-Berne potential
6. in.intel.eam - Copper benchmark with Embedded Atom Method
7. in.intel.sw - Silicon benchmark with Stillinger-Weber
8. in.intel.tersoff - Silicon benchmark with Tersoff
9. in.intel.water - Coarse-grain water benchmark using Stillinger-Weber # #################

2. Expected Timesteps/second with turbo on and HT enabled, LAMMPS June-2017
3. - Compiled w/ Intel Parallel Studio 2017u2 and Makefile.intel_cpu_intelmpi #
4. Xeon E5-2697v4 Xeon Phi 7250
6. in.intel.lj - 199.5 282.3
7. in.intel.rhodo - 12.4 17.5
8. in.intel.lc - 19.0 25.7
9. in.intel.eam - 59.4 92.8
10. in.intel.sw - 132.4 161.9
11. in.intel.tersoff - 83.3 101.1
12. in.intel.water - 53.4 90.3 # #################

2. For Skylake server (Xeon) architectures, see notes in the USER-INTEL/README
3. for build flags that should be used. #################

2. For Haswell (Xeon v3) architectures, depending on the compiler version,
3. it may give better performance to compile for an AVX target (with -xAVX
4. compiler option) instead of -xHost or -xCORE-AVX2 for some of the
5. workloads. In most cases, FMA sensitive routines will still use AVX2
6. (MKL and SVML detect the processor at runtime). For Broadwell (Xeon v4)
7. architectures, -xCORE-AVX2 or -xHost will work best for all. #################

2. The default benchmark timesteps will run between 30s and 1 minute with
3. the Intel package. You can specify a multiplier for all of the benchmarks
4. to increase or decrease the runtime. Example commandline arguments: #
5. -v m 2 # Run for twice as long
6. -v m 0.5 # Run for half as long #################

2. The LAMMPS newton setting can be controlled from the commandline for the
3. benchmarks with the N variable: #
4. -v N on # newton on
5. -v N off # newton off #
6. The default is on for all of the benchmarks except for LJ where the off
7. setting performs best with the USER-INTEL package #################

# Example for running benchmarks (see run_benchmarks.sh for script):

Number of physical cores per node not including hyperthreads

export LMP_CORES=28

1. If hyperthreading is enabled, number of hyperthreads to use per core
2. (2 for Xeon; 2 or 4 for Xeon Phi)

export OMP_NUM_THREADS=2

Name of the LAMMPS binary

export LMP_BIN=../../lmp_intel_cpu

LAMMPS root directory

export LMP_ROOT=../../../

source source /opt/intel/parallel_studio_xe_2017.2.050/psxevars.sh export KMP_BLOCKTIME=0 export I_MPI_PIN_DOMAIN=core export I_MPI_FABRICS=shm # For single node

ONLY FOR INTEL XEON PHI x200 SERIES PROCESSORS

export I_MPI_SHM_LMT=shm

Generate the restart file for use with liquid crystal benchmark

mpirun -np $LMP_CORES $LMP_BIN -in in.lc_generate_restart -log none

Benchmark to run

export bench=in.intel.lj

2. For Intel Xeon Phi x200 series processors best performance is achieved by
3. using MCDRAM. In flat mode, this can be achieved with numactl,
4. MPI environment variables, or other options provided by batch schedulers #################

2. To run without a optimization package #################

mpirun -np $LMP_CORES $LMP_BIN -in $bench -log none -v N on

2. To run with USER-OMP package #################

mpirun -np $LMP_CORES $LMP_BIN -in $bench -log none -pk omp 0 -sf omp -v N on

2. To run with USER-INTEL package and no coprocessor #################

mpirun -np $LMP_CORES $LMP_BIN -in $bench -log none -pk intel 0 -sf intel

2. To run with USER-INTEL and automatic load balancing to 1 coprocessor #################

mpirun -np $LMP_CORES $LMP_BIN -in $bench -log none -pk intel 1 -sf intel

2. If using PPPM (e.g. in.intel.rhodo) on Intel Xeon Phi x200 series
3. or Skylake processors #################

export KMP_AFFINITY=none rthreads=$((OMP_NUM_THREADS-1)) mpirun -np $LMP_CORES $LMP_BIN -in $bench -log none -pk intel 0 omp $rthreads lrt yes -sf intel