Tensorflow-gpu installation issue on ubuntu16.04

Question

I have a linux machine running Ubuntu16.04, NVIDIA GEFORCE1060. I am running python3.6 (installed using anaconda) and CUDA9.1

I was happily coding in tensorlow with CPU with no issues. Then I wanted to install tensorflow-gpu and that's where I am having issues.

First installed cudnn7.1.2 successfully ( make and run mnist-cudnn completed successfully)

pip3 install tensorflow_gpu-1.7.0-cp36-cp36m-manylinux1_x86_64.whl

---

import tensorflow as tf

I am getting the following errors

Traceback (most recent call last):
  File "/home/bony/anaconda3/lib/python3.6/site-packages/tensorflow/python/pywrap_tensorflow.py", line 58, in <module>
    from tensorflow.python.pywrap_tensorflow_internal import *
  File "/home/bony/anaconda3/lib/python3.6/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 28, in <module>
    _pywrap_tensorflow_internal = swig_import_helper()
  File "/home/bony/anaconda3/lib/python3.6/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 24, in swig_import_helper
    _mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description)
  File "/home/bony/anaconda3/lib/python3.6/imp.py", line 243, in load_module
    return load_dynamic(name, filename, file)
  File "/home/bony/anaconda3/lib/python3.6/imp.py", line 343, in load_dynamic
    return _load(spec)
ImportError: libcublas.so.9.0: cannot open shared object file: No such file or directory
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/bony/anaconda3/lib/python3.6/site-packages/tensorflow/__init__.py", line 24, in <module>
    from tensorflow.python import *  # pylint: disable=redefined-builtin
  File "/home/bony/anaconda3/lib/python3.6/site-packages/tensorflow/python/__init__.py", line 49, in <module>
    from tensorflow.python import pywrap_tensorflow
  File "/home/bony/anaconda3/lib/python3.6/site-packages/tensorflow/python/pywrap_tensorflow.py", line 74, in <module>
    raise ImportError(msg)
ImportError: Traceback (most recent call last):
  File "/home/bony/anaconda3/lib/python3.6/site-packages/tensorflow/python/pywrap_tensorflow.py", line 58, in <module>
    from tensorflow.python.pywrap_tensorflow_internal import *
  File "/home/bony/anaconda3/lib/python3.6/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 28, in <module>
    _pywrap_tensorflow_internal = swig_import_helper()
  File "/home/bony/anaconda3/lib/python3.6/site-packages/tensorflow/python/pywrap_tensorflow_internal.py", line 24, in swig_import_helper
    _mod = imp.load_module('_pywrap_tensorflow_internal', fp, pathname, description)
  File "/home/bony/anaconda3/lib/python3.6/imp.py", line 243, in load_module
    return load_dynamic(name, filename, file)
  File "/home/bony/anaconda3/lib/python3.6/imp.py", line 343, in load_dynamic
    return _load(spec)
ImportError: libcublas.so.9.0: cannot open shared object file: No such file or directory
Failed to load the native TensorFlow runtime.

Any help to resolve this and get me going on tensorflow-gpu will be appreciated

Answer 1

• Check this github issue https://github.com/tensorflow/tensorflow/issues/15604
• You said you have installed cude9.1 but it is still showing that you have cuda9.0
• And check whether tensor-flow version you are using is compatible with cuda9.1 or not because many of the comments say that tensor-flow expects cuda9.0 and not cuda9.1
• If you are using anaconda then you can use the following command:
  1] conda install -c anaconda tensorflow-gpu
  2] conda install -c aaronzs tensorflow-gpu
• I think the above command also installs the cuda-toolkit and does all the required configuration for you which makes your life easier

Answer 2

• I have the same issue with you on Ubuntu 16.04
• This my solution

• First,on the official web page part of NVIDIA requirements to run TensorFlow with GPU support we can know :

  • CUDA® Toolkit 9.0.
  • cuDNN v7.0.
  • GPU card with CUDA Compute Capability 3.0 or higher for building from source and 3.5 or higher for our binaries

• So uninstall Cuda9.1.

  • If you do not know how to uninstall,you can install Synaptic Package Manager in Ubuntu Software,and Search cuda and select Mark for complete removal
  • Apply
  • Reload Synaptic Package Manager
  • Select Cuda-9.0 Mark for install
  • Apply

• Install cnDNN v7.0

  • I had download cuDNN v7.0.5 Developer Library for Ubuntu16.04 (Deb) ,because it is the latest version cuDNN v7.0 for Cuda-9.0

• Install cuda-command-line-tools

  sudo apt-get install cuda-command-line-tools

  • If can not find cuda-command-line-tools

    sudo apt-cache search cuda-command-line-tool

    And select install cuda-command-line-tool-9.0

• Follow official document,add its path to your LD_LIBRARY_PATH environment variable

  export LD_LIBRARY_PATH=${LD_LIBRARY_PATH:+${LD_LIBRARY_PATH}:}/usr/local/cuda/extras/CUPTI/lib64

• I am using Anaconda3, i use conda command create environment include py3.5 name py35.

• So,i use source activate py35 and use pip install --upgrade https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-1.7.0-cp35-cp35m-linux_x86_64.whl .

• You can select tensorflow-gpu which one support you python version.

• Validate your installation

  • Run a short TensorFlow program import tensorflow as tf hello = tf.constant('Hello, TensorFlow!') sess = tf.Session() print(sess.run(hello))
  • If the system outputs Hello, TensorFlow! , then you are ready to begin writing TensorFlow programs.

Good luck to you.