2

I am trying to install tensorflow-gpu on my macbook pro and have tried pretty much everything which I will briefly mention. But first here is my setup Macbook pro retina 15" High Sierra 10.13.4 NVIDIA GT650M card

Over the past two weeks I have tried all sorts of combinations and fed up with drivers/versions especially with High Sierra 10.13.4 and variables and so on.

If anybody has had success with this please could they tell me or point me to the versions and method for CUDA CUDNN Tensorflow-gpu (understand 1.1 is the highest for mac) XCode (I have 9.2)

and anything else.

I have my anaconda environment working well for all of the machine learning stuff on CPU and consider using the GPU to be the next challenge.

a-robot
  • 37
  • 1
  • 2

2 Answers2

2

Here's a link that may help.

I have installed using the same guidelines and it's working for me on macbook pro with same configuration.

Nikhil Ranjan
  • 185
  • 1
  • 6
2

I was able to compile tensorflow 1.4.0 with GPU support for my MacBook Pro (Retina, 15-inch, Late 2013) with GT 750M card under MacOS 10.12.6 with instructions from the following link:

http://paolino.me/tutorial/tensorflow/machine-learning/deep-learning/gpu/2017/11/18/installing-tensorflow-1.4.0-macos-cuda/

giminey
  • 126
  • 1
  • 3
  • Thank you. So far I have installed CUDA/cudnn/xcode and got the devQuery bandwidth CUDA samples to work (yaaay!) - I am struggling with the bazel tensorflow bit. Thank you once again. Will keep updating when I make progress. I have cuda driver 396.64 cuda 9.2 cudnn v7.1. One question do I have to disable SIP every time I restart my MacBook? – a-robot Jun 19 '18 at 06:56
  • "Compiling TensorFlow can be time-consuming, so it's worth trying a pre-built package first. I have uploaded the wheel files I built to Google Drive, https://drive.google.com/drive/folders/1tvNVxbY9Bc2RRVK3wokxs0X88S-x04qr?usp=sharingso download the relevant file, activate the relevant Python environment, and $ pip install /path/to/tensorflow-1.6.0-cp36-cp36m-macosx_10_13_x86_64.whl # Python 3.6 $ pip install /path/to/tensorflow-1.6.0-cp27-cp27m-macosx_10_13_x86_64.whl # Python 2.7" – giminey Jun 20 '18 at 16:32
  • you don't need to disable SIP everytime, just one time – giminey Jun 20 '18 at 16:34
  • Thank you once again.Will try again. I spent a lot of time clearing out anaconda and reinstalling python 3 and used [link](http://paolino.me/tutorial/tensorflow/machine-learning/deep-learning/gpu/2017/11/18/installing-tensorflow-1.4.0-macos-cuda/) and configured tensorflow and got to the bazel build which is where I stumbled. it kept saying the version was higher. And then I hit upon the workspace not available error. On the plus side my CUDA still works! For the relevant python env do you mean Python 3? Can I use one of my machine learning envs which has Python 3.5? Thank you very much – a-robot Jun 21 '18 at 06:58
  • you may need to upgrade or downgrade your bazel version. if you're installing from the instructions you linked, you should be using bazel 0.7.0. I couldn't get tensorflow 1.4 to build with higher bazel versions. – giminey Jun 21 '18 at 20:44