I am on a docker image, so I can not access the "outside" of the docker image. I want to install tensorflow with gpu support so used:
pip install tensorflow-gpu
cudnn and CUDA is installed and working. An old version (0.11) is available in the image and is running with CUDA and cudnn just fine, but I need to upgrade to version 1 or higher. I have two Nvidia Titans.
After using the shown pip command I use the following script to see if I have GPU support enabled, and also look at nvidia-smi:
import tensorflow as tf
# Creates a graph.
with tf.device('/gpu:0'):
a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='a')
b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='b')
c = tf.matmul(a, b)
# Creates a session with log_device_placement set to True.
sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))
# Runs the op.
print sess.run(c)
After this I only get the output
No module named tensorflow
If I check the pip list with:
pip list | grep tensorflow
I get the output:
tensorflow-gpu (1.0.1)
Is it a simple wrong import?
If I use the non-gpu-support install pip install tensorflow the above code gives:
Device mapping: no known devices.
Which of course is due to no support of the gpu. So to summarize, how do I get the GPU Version of tensorflow to work with a simple pip install and a version above 1.0 ?
Installing it with
conda install tensorflow-gpu
resolved all issues.