How do I make sure my Keras/tensorflow code uses my MacBook Pro's AMD graphics card

Question

I am running some Keras/tensorflow code in python on my MacBook Pro with Radeon Pro 560X 4096 MB and Intel UHD Graphics 630 1536 MB. What do I have to do to use the graphics cards in running the neural network code?

You would need to use this version of tensorflow, if it supports your GPU: https://github.com/ROCmSoftwarePlatform/tensorflow-upstream — Dr. Snoopy, Mar 04 '19 at 10:25

score 3 · Answer 1 · edited Apr 06 '19 at 08:09

3

If you are running Keras, then you can use PlaidML as a backend: https://github.com/plaidml/plaidml

Installation is as easy as:

virtualenv plaidml
source plaidml/bin/activate
pip install plaidml-keras plaidbench

Then you need to run setup (to select default processor):

plaidml-setup

Then you need to add:

import os
os.environ["KERAS_BACKEND"] = "plaidml.keras.backend"

At the beginning of your script.

Best

edited Apr 06 '19 at 08:09

marc_s

732,580
175
1,330
1,459

answered Mar 19 '19 at 09:33

Paweł D

69
6

is it a must to use virtualenv? if i don't use virtualenv, does this affect my other python programming? – yts61 Nov 05 '20 at 11:02
no it isn't. you can install it in your regular environment – Jürgen K. Oct 25 '21 at 11:50

RaJa · Answer 2 · 2019-03-19T10:07:11.840

0

If you want to stick to Tensorflow then you need to compile TensorFlow on your own machine so that it works with OpenCL. Otherwise, it will only work with CUDA which is restricted to NVidia.

Using Keras & Tensorflow with AMD GPU

edited Mar 19 '19 at 10:07

answered Mar 19 '19 at 09:57

RaJa

1,471
13
17

How do I make sure my Keras/tensorflow code uses my MacBook Pro's AMD graphics card

2 Answers2