Tensorflow doesn't use GPU

Accelerated Computing CUDA CUDA Setup and Installation
1

PROBLEM:
In python tf.config.list_physical_devices(‘GPU’) doesn’t find GPU.
(However CUDA samples in C:\ProgramData\NVIDIA Corporation\CUDA Samples\v10.2\5_Simulations run successfully.)

INFORMATION:
Intel i7-6700 CPU. 32 GB RAM.
Windows 10. GeForce GTX 1070. GeForce Game Ready Driver 442.19

Previously, CUDA and Tensorflow worked with GPU. After driver update for approx a year ago it doesn’t any more.

Recently I have:
Updated NVIDIA driver.
Uninstalled Tensorflow, python 3.7, cuDNN, CUDA, VIsual Studio.
Installed Visual Studio.
Installed CUDA 10.2 [cuda_10.2.89_441.22_win10].
Made sure the CUDA samples (in C:\ProgramData\NVIDIA Corporation\CUDA Samples\v10.2\5_Simulations) compile and Run in Visual Studio.
Installed cuDNN (and moved the .dll, .h, .lib files) [cudnn-10.2-windows10-x64-v7.6.5.32.zip]
Installed python [python-3.7.6-amd64.exe]
Installed Tensorflow using pip3

PROBLEM IN PYTHON ON WINDOWS:
Python 3.7.6 (tags/v3.7.6:43364a7ae0, Dec 19 2019, 00:42:30) [MSC v.1916 64 bit (AMD64)] on win32
Type “help”, “copyright”, “credits” or “license()” for more information.

import tensorflow as tf
tf.test.is_built_with_cuda()
True
tf.config.list_physical_devices(‘GPU’)

Any ideas - or missing information?
Thanks.
/Stefan

2

https://www.tensorflow.org/install/gpu

3

I wonder why I did not notice that … uninstalled and installed CUDA 10.1 etc … TF is using my GPU now. You are a star - thanks for the quick answer!

4

PROBLEM:

I have a similar problem. I have done all the steps discussed above but with CUDA 10.1.

“from tensorflow.python.client import device_lib
…”

shows that tf can see my GPU.

tf.test.is_GPU_avalible - returns true

However, when I run a script it does not utilise the GPU.

I have been stuck trying to fix this for days, any help would be massively appreciated.