Running on PNNL Systems

Table of Contents

1. Puma
2. DGX-1
3. Constance

Puma

DGX-1

The installation path for DGX-1 is /raid/matex/0.6/

To use MaTEx TensorFlow using GPUs 0,1,2 and 3, run

source /raid/matex/0.6/run_TFEnv_DGX1.sh
export CUDA_VISIBLE_DEVICES=0,1,2,3

Example scripts are located in

/raid/matex/0.6/examples/glibc_before_2.23/

To run a python script example.py use the following code after sourcing run_TFEnv_DGX1.sh:

mpirun -n 4 --mca opal_event_include poll $FAKE_SYSTEM_LIBS/lib/x86_64-linux-gnu/ld-linux-x86-64.so.2 --library-path $PNETCDF_INSTALL_DIR/lib:$FAKE_SYSTEM_LIBS/lib/:$FAKE_SYSTEM_LIBS/lib/x86_64-linux-gnu/:$FAKE_SYSTEM_LIBS/usr/lib64/gconv:$FAKE_SYSTEM_LIBS/usr/lib64/audit:$LD_LIBRARY_PATH $PYTHONHOME/bin/python $PWD/example.py

Constance (as of 6/13/2017)

The installation path for Constance is /people/sieg052/matex/src/deeplearning/tensorflow/cpu/py3.x/

MaTEx TensorFlow is currently only configured for CPUs on Constance. To use set up the environment, run

source /people/sieg052/env_matex_cpu.sh
source /people/sieg052/matex/src/deeplearning/tensorflow/cpu/py3.x/run_TFEnv.sh

Example scripts are located in

/people/sieg052/matex/src/deeplearning/tensorflow/examples/glibc_before_2.23

To run a python script example.py use the following code after sourcing run_TFEnv.sh and setAlias.sh:

mpirun -n $nodes --mca opal_event_include poll $FAKE_SYSTEM_LIBS/lib/x86_64-linux-gnu/ld-linux-x86-64.so.2 --library-path $PNETCDF_INSTALL_DIR/lib:$FAKE_SYSTEM_LIBS/lib/:$FAKE_SYSTEM_LIBS/lib/x86_64-linux-gnu/:$FAKE_SYSTEM_LIBS/usr/lib64/gconv:$FAKE_SYSTEM_LIBS/usr/lib64/audit:$LD_LIBRARY_PATH $PYTHONHOME/bin/python $PWD/example.py