Tensorflow not using GPU with jupyter (Windows)

Question

I have tensorflow GPU working with CUDA.

But when I am running the following code in jupyter notebook:

import sys
import numpy as np
import tensorflow as tf
from datetime import datetime

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 allow_soft_placement and log_device_placement set
# to True.
sess = tf.Session(config=tf.ConfigProto(
      allow_soft_placement=True, log_device_placement=True))
# Runs the op.
print (sess.run(c))

The command prompt shows that CPU was used and not the GPU.

Command Prompt Screen

Are you sure your Jupyter notebook is using the correct anaconda environment (tensorflow-gpu2)? I have seen it where it uses some other default environment and one needs to change that — Luca
I think there should be a conda tab and at least it will tell you which one is the default one. For the moment, I would change that and see if that changes stuff. I am not near a dev environment now but you might wanna google on how you can launch the notebook to use a particular python env. — Luca
I was using the right virtualenv. Everything was working fine until I tried to use CPU version of Tensorflow. Earlier TF-GPU was working. I think this answers the question: stackoverflow.com/questions/42473052/… — rahulmanuwas

BG Park BG Park · Accepted Answer · 2017-03-07T14:52:18

Please Try Nightly Build version. It seems like you are using 1.0 official release version.

$ pip install --ignore-installed --upgrade https://storage.googleapis.com/tensorflow/windows/gpu/tensorflow_gpu-1.0.0-cp35-cp35m-win_amd64.whl

You can find latest wheel pkg url from https://github.com/tensorflow/tensorflow#installation

Tensorflow not using GPU with jupyter (Windows)

1 Answers