Tensorflow cannot use GPU

Question

I have installed CUDA because I need to run some Deep Neural Network models, but Tensorflow is still unable to see the gpu

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

It shows only the CPU.

How can I get Tensorflow to use my gpu?

error log I get when I import keras with tensorflow-gpu installed ImportError: Traceback (most recent call last): File "C:\Users\Monviso\Anaconda3\lib\site-packages\tensorflow\python\pywrap_tensorflow_internal.py", line 18, in swig_import_helper return importlib.import_module(mname) File "C:\Users\Monviso\Anaconda3\lib\importlib__init__.py", line 126, in import_module return _bootstrap._gcd_import(name[level:], package, level) File "", line 978, in _gcd_import File "", line 961, in _find_and_load File "", line 950, in _find_and_load_unlocked File "", line 648, in _load_unlocked File "", line 560, in module_from_spec File "", line 922, in create_module File "", line 205, in _call_with_frames_removed ImportError: DLL load failed: The specified module could not be found.

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "C:\Users\Monviso\Anaconda3\lib\site-packages\tensorflow\python\pywrap_tensorflow.py", line 41, in <module>
    from tensorflow.python.pywrap_tensorflow_internal import *
  File "C:\Users\Monviso\Anaconda3\lib\site-packages\tensorflow\python\pywrap_tensorflow_internal.py", line 21, in <module>
    _pywrap_tensorflow_internal = swig_import_helper()
  File "C:\Users\Monviso\Anaconda3\lib\site-packages\tensorflow\python\pywrap_tensorflow_internal.py", line 20, in swig_import_helper
    return importlib.import_module('_pywrap_tensorflow_internal')
  File "C:\Users\Monviso\Anaconda3\lib\importlib\__init__.py", line 126, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
ModuleNotFoundError: No module named '_pywrap_tensorflow_internal'


Failed to load the native TensorFlow runtime.

See https://www.tensorflow.org/install/install_sources#common_installation_problems

for some common reasons and solutions.  Include the entire stack trace
above this error message when asking for help.

What versions of Nvidia driver and cudNN do you have? What model of video card do you have installed? — Maxim Norin
+1 to deaspo's comment. There are two versions of tensorflow (tensorflow and tensorflow-gpu, if I recall it right) — Maxim Norin
I have installed both tensorflow version. My gpu is a gtx960M , therefore compatible, I have CUDA 9.0 — Alex
@Alex: You need to remove the tensorflow CPU version to access the GPU version of tensorflow. — Beta

Abhijit Balaji Abhijit Balaji · Accepted Answer · 2018-02-03T04:26:08

Your Cuda version is the problem. Only the tensorflow version 1.5 supports CUDA 9 and cuDNN 7. The previous versions of the Tensorflows support only CUDA 8 and cuDNN 6. If you have installed using anaconda it is very likely that you have not installed version 1.5. Use pip or pip3 to install.

Since you are using a windows machine check this link to install tensorflow with gpu support.

Check this link to see which version require what Cuda and Cudnn

I recommend you to uninstall both the CPU and GPU version of tensorflow and install afresh.

If you wanna keep your cuda version to 9 then install tensorflow version 1.5. If you install using anaconda it is likely you will get version 1.4 which supports only Cuda 8

Tensorflow cannot use GPU

1 Answers