Tensorflow-gpu only running on CPU

Question

I am running tensorflow 1.8.0 installed by 'pip -install tensorflow-gpu' with CUDA 9.0 and with cuDNN 7.1.4 on Windows 10 and I have been trying to get tensorflow to work with my NVIDIA GeForce GTX 860M graphics card. I am using IDLE to run and write my code in.

When I run any code it only performs on my CPU. Some sample code im running until I can get it to run on my gpu to do the MNIST tensorflow tutorials is:

    import tensorflow as tf


    # Initialize two constants
    x1 = tf.constant([1,2,3,4])
    x2 = tf.constant([5,6,7,8])

    # Multiply
    result = tf.multiply(x1, x2) 

    # Intialize the Session
    sess = tf.Session()

    # Print the result
    print(sess.run(result))

    # Close the session
    sess.close()

When i run this I get this failed to create session error:

    tensorflow.python.framework.errors_impl.InternalError: Failed to create session.

So I added the following lines to the beginning of the code

     import os
     os.environ["CUDA_VISIBLE_DEVICES"] = '1'

and now the code runs succesfully with the correct output

   [ 5 12 21 32]

but it only runs on my CPU, and when I try the following commands in the IDLE command line,

    from tensorflow.python.client import device_lib
    device_lib.list_local_devices()

the output is

    [name: "/device:CPU:0"
    device_type: "CPU"
    memory_limit: 268435456
    locality {
    }
    incarnation: 1631189080169918107
    ]

I have tried restarting my computer, unistalling and reinstalling tensorflow-gpu and deleted other versions of CUDA I previously had installed, but I can not get my program to recognize my NVIDIA graphics card, and I looked it up and it is a supported card.

What else can I do to get my code to recognize and run on my graphics card?

Your GPU came in two revisions, one has a CUDA capability level of 3.0, the other 5.0. AFAIR tensorflow-gpu as Windows binary requires 3.5. — Klaus D.
– Klaus D., Commented Jun 3, 2018 at 23:07
I tried running deviceQuery to check which one mine was but it said: CUDA driver version is insufficient for CUDA runtime version. Does this mean that mine is the 3.0 version? — Larr Bear
– Larr Bear, Commented Jun 4, 2018 at 0:00
After some digging in my PC i found out I have the 3.0 older kepler version of the 860M, Im going to delete everything and build from source with instructions from github.com/tensorflow/tensorflow/issues/6001 — Larr Bear
– Larr Bear, Commented Jun 4, 2018 at 0:29

pranshu · Accepted Answer · 2018-06-03 23:04:46Z

1

Installing CUDA is not enough, you also need to install cuDNN 7.0.5 and some other dependencies too. Refer http://www.python36.com/install-tensorflow-gpu-windows/ for complete detail.

answered Jun 3, 2018 at 23:04

pranshu

1331 silver badge8 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

Tensorflow-gpu only running on CPU

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related