Current as of Sept 6, 2017
I have a MacBook Pro Retina 2013 and I had a hard time setting up TensorFlow with GPU support from all the conflicting information out there. I hope this gets you up and running training models on your local computer!
- Download the CUDA Toolkit 8.0 for macOS here. Install it using the DMG package.
- Edit your ~/.bash_profile file and add the following line:
export DYLD_LIBRARY_PATH="/Developer/NVIDIA/CUDA-8.0/lib:$DYLD_LIBRARY_PATH"
- Download the CUDA DNN v5.1 from here. This requires you to sign up as an NVIDIA Developer.
- Unzip the CUDA DNN package and copy the files inside the
include
folder toDeveloper/NVIDIA/CUDA-8.0/include
and copy the files inside thelib
folder intoDeveloper/NVIDIA/CUDA-8.0/lib
- Inside your terminal, do
source ~/.bash_profile
- Inside your Python environment, do
pip install tensorflow-gpu
. If you have tensorflow (regular CPU version) already installed, you will need to uninstall it first withpip uninstall tensorflow
. If you aren't using an Virtual Environment to keep you python stuff separate, I highly recommend it (check it out here. - Confirm your installation by entering
python -c "import tensorflow"
in your terminal.
That's all!
this installs legacy tensorflow 1.1