Benchmarking: Amber 16 with GPU v100

• Note: Centos 7
• Amber 16 with Amber tools 17

Software

[david@head-Boston benchmarks]$ pwd
/home/david/benchmarks
[david@head-Boston benchmarks]$ ls amber16
Amber16_Benchmark_Suite.tar.bz2  Amber16.tar.bz2  AmberTools17.tar.bz2

PreReqs

sudo yum install gcc gcc-gfortran gcc-c++ flex tcsh zlib-devel \
          bzip2-devel libXt-devel libXext-devel libXdmcp-devel tkinter \
          openmpi openmpi-devel perl perl-ExtUtils-MakeMaker patch bison

# Make sure to setup environment 
[david@sanger_4xv100 amber]$ tail -n 2 ~/.bashrc
export PATH=/usr/lib64/openmpi/bin/:$PATH
export LD_LIBRARY_PATH=/usr/lib64/openmpi/lib/:$LD_LIBRARY_PATH

# check mpi for cuda support 
[david@sanger_4xv100 amber]$ ompi_info --parsable --all | grep mpi_built_with_cuda_support:value
mca:mpi:base:param:mpi_built_with_cuda_support:value:false

Unpack

tar jxvf  AmberTools17.tar.bz2
tar jxvf  Amber16.tar.bz2
cd amber16

Build

export AMBERHOME=/home/david/amber/amber16
export AMBER_PREFIX=/home/david/amber/amber16
export AMBER_SOURCE=/home/david/amber/amber16
export CUDA_HOME=/usr/local/cuda-9.0

# If you wish to run both single and multiple-GPU runs, you must perform the configure and build steps for both with MPI and without. 

./configure -cuda -noX11 gnu
make install -j32
./configure -cuda -mpi -noX11 gnu
make install -j32