Tensorflow can't find GPU

1
votes

My apologies as there's a million questions out there like this, but none of them seem to answer mine. I'm trying to re-install Tensorflow so that it uses my GPU.

I'm running:

  • Windows 10
  • Python 3.6.5
  • Keras 2.2.4
  • TF 1.13.1
  • Nvidia Quadro M1000M (driver 412.16, compute capability 5.0)
  • CUDA 10.0.130
  • Have CUPTI installed and CuDNN files copied into CUDA folders.

The CUDA v10.0 folder, the \extras\CUPTI\libx64 and the \include folder are all in my PATH, also CUDAPATH is as well as CUDA_PATH is defined)

Yet, with pip install tensorflow it only finds my CPU (using:

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

And I can't seem to be able to import tensorflow if I 

pip uninstall tensorflow
pip install tensorflow-gpu
python
import tensorflow

ModuleNotFoundError: No module named 'tensorflow'
1
pythontensorflowgpu
Have you tried restarting your computer since setting your PATH variables? They do not completely set in until you do a clean restart of your computer. - Snel23
Found an answer here after all: Looks like some leftover modules weren't properly uninstalled. stackoverflow.com/questions/42326748/… - Niels Uitterdijk
Possible duplicate of tensorflow on GPU: no known devices, despite cuda's deviceQuery returning a "PASS" result - Niels Uitterdijk
sometimes you can have more than one version of TF installed, you have to keep using pup uninstall tensorflow until they are all gone, then use pip install tensorflow-gpu - Andrew Louw

1 Answers

1
votes

Try installing Tensorflow again with option --ignore-installed such as:

pip install tensorflow-gpu==2.0.0-alpha0 --ignore-installed