Sunday, June 8, 2014

Uncurated notes on Gromacs GPU installation

This file is in /scratch [flute]

GPU installation on Spear was successful with this cmake

cmake .. -DGMX_GPU=ON -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda -DGMX_BUILD_OWN_FFTW=ON

[Performance]
12102 atoms CG

Using 2 MPI threads
Using 6 OpenMP threads per tMPI thread
Detecting CPU-specific acceleration.
Present hardware specification:
Vendor: AuthenticAMD
Brand:  AMD Opteron(tm) Processor 4184
Family: 16  Model:  8  Stepping:  1
Features: apic clfsh cmov cx8 cx16 htt lahf_lm misalignsse mmx msr nonstop_tsc pdpe1gb popcnt pse rdtscp sse2 sse3 sse4a
Acceleration most likely to fit this hardware: SSE2
Acceleration selected at GROMACS compile time: SSE2


2 GPUs detected:
  #0: NVIDIA Tesla M2050, compute cap.: 2.0, ECC: yes, stat: compatible
  #1: NVIDIA Tesla M2050, compute cap.: 2.0, ECC: yes, stat: compatible

2 GPUs auto-selected for this run.
Mapping of GPUs to the 2 PP ranks in this node: #0, #1

Cut-off's:   NS: 1.441   Coulomb: 1.2   LJ: 1.2
               Core t (s)   Wall t (s)        (%)
       Time:    48384.100     4035.993     1198.8
                         1h07:15
                 (ns/day)    (hour/ns)
Performance:     2140.737        0.011 <----------- font="">
Finished mdrun on node 0 Thu Mar 20 18:40:36 2014

------------ [CPU Edison install] ----------------
git clone git://git.gromacs.org/gromacs.git
module purge
module load intel
module load cmake
module load openmpi-ccm
module load fftw
mkdir 1049pm-mar24-build
export CCDIR=/opt/intel/composer_xe_2013_sp1.0.080/bin/intel64/
export MPICCDIR=/usr/common/usg/openmpi-ccm/1.6.5/intel/bin/
export FFTW_LOCATION=/opt/fftw/3.3.0.4/sandybridge/
export CXX=mpicxx
export CC=mpicc
mkdir ~/gromacs/1049pm-mar24-build/
cmake ../   -DFFTW3F_INCLUDE_DIR=$FFTW_LOCATION/include       -DFFTW3F_LIBRARIES=$FFTW_LOCATION/lib/libfftw3f.a       -DCMAKE_INSTALL_PREFIX=/global/homes/p/pavang/gromacs/1049pm-mar24-build/       -DGMX_X11=OFF       -DCMAKE_CXX_COMPILER=${MPICCDIR}/mpicxx       -DCMAKE_C_COMPILER=${MPICCDIR}/mpicc       -DGMX_MPI=ON       -DGMX_PREFER_STATIC_LIBS=ON -DFFTWF_LIBRARY=$FFTW_LOCATION/lib/libfftw3f.a -DFFTWF_INCLUDE_DIR=$FFTW_LOCATION/include

make
make install
source /global/homes/p/pavang/gromacs/1049pm-mar24-build/bin/GMXRC.bash 
mdrun_mpi -v -deffnm  em

[Performance]
38315 atoms Lysozyme in water

On 1 MPI rank, each using 32 OpenMP threads
               Core t (s)   Wall t (s)        (%)
       Time:     7596.371      252.435     3009.2
                 (ns/day)    (hour/ns)
Performance:       11.802        2.034 <-------------- font="">

No comments:

Post a Comment