Optimized_TF

Install instructions for GPU use

1. Navigate to your Eagle home directory or scratch directory using

cd ~/

or

cd /scratch/$USER/

2. Clone the github repo into the directory you chose

git clone https://github.com/NREL/HPC

3. Navigate to the repo

cd ./HPC/general/Optimized_TF/

4. To install TensorFlow 2.4.0 with Python 3.8 for GPUS run the following

• a) load the appropriate modules

  module purge
  module use /nopt/nrel/apps/modules/test/modulefiles/
  module load conda
  module load gcc/7.4.0
  module load cudnn/8.0.5/cuda-10.2
  

• b) build a predefined conda environment

  conda env create -f py38tf24.yml
  

• c) Active the conda environment

  source activate py38tf24
  

• d) Install the precompiled TensorFlow installation from a wheel

  pip install --upgrade --no-deps --force-reinstall /nopt/nrel/apps/wheels/tensorflow-2.4.0-cp38-cp38-linux_x86_64.whl
  

• e) If you are on an allocated or interactive node you can test the install by running

  python3 -c "import tensorflow as tf; print(tf.reduce_sum(tf.random.normal([1000, 1000])))"
  

5. To install TensorFlow 2.3.2 with Python 3.8 for GPUS run the following

• a) load the appropriate modules

  module purge
  module use /nopt/nrel/apps/modules/centos74/modulefiles/
  module load gcc/7.4.0
  module load cuda/10.0.130
  module load cudnn/7.4.2/cuda-10.0
  module load conda
  

• b) build a predefined conda environment

  conda env create -f py38tf23.yml
  

• c) Activate the conda environment

  source activate py38tf23
  

• d) Install the precompiled TensorFlow installation from a wheel

  pip install --upgrade --no-deps --force-reinstall /nopt/nrel/apps/wheels/tensorflow-2.3.2-cp38-cp38-linux_x86_64.whl
  

• e) If you are on an allocated or interactive node you can test the install by running

  python3 -c "import tensorflow as tf; print(tf.reduce_sum(tf.random.normal([1000, 1000])))"
  

6. To install TensorFlow 2.2.x with Python 3.7 for GPUs run the following

• a) load the appropriate modules

  module purge
  module use /nopt/nrel/apps/modules/centos74/modulefiles/
  module load gcc/7.4.0
  module load cuda/10.0.130
  module load cudnn/7.4.2/cuda-10.0
  module load conda
  

• b) build a predefined conda environment

  conda env create -f py37tf22.yml
  

• c) Active the conda environment

  source activate py37tf22
  

• d) Install the precompiled TensorFlow installation from a wheel

  pip install --upgrade --no-deps --force-reinstall /nopt/nrel/apps/wheels/tensorflow-2.2.1-cp37-cp37m-linux_x86_64.whl
  

  or

  pip install --upgrade --no-deps --force-reinstall /nopt/nrel/apps/wheels/tensorflow-2.2.1-cp37-cp37m-linux_x86_64.whl
  

• e) If you are on an allocated or interactive node you can test the install by running

  python3 -c "import tensorflow as tf; print(tf.reduce_sum(tf.random.normal([1000, 1000])))"
  

7. To install TensorFlow 2.0.0 with Python 3.7 for GPUs run the following

• a) load the appropriate modules

  module purge
  module use /nopt/nrel/apps/modules/centos74/modulefiles/
  module load gcc/7.3.0
  module load cuda/10.0.130
  module load cudnn/7.4.2/cuda-10.0
  module load conda
  

• b) build a predefined conda environment

  conda env create -f py37tf20.yml
  

• c) Active the conda environment

  source activate py37tf20
  

• d) Install the precompiled TensorFlow installation from a wheel

  pip install --upgrade --no-deps --force-reinstall /nopt/nrel/apps/wheels/tensorflow-2.0.0-cp37-cp37m-linux_x86_64.whl
  

• e) If you are on an allocated or interactive node you can test the install by running

  python3 -c "import tensorflow as tf; print(tf.reduce_sum(tf.random.normal([1000, 1000])))"