Trouble activating Tensorflow 2.0 on GPU

Question

This has been asked countless times here, but I could not find solution that fits my problem. I am trying to install tensorflow-gpu on Windows 10, Anaconda. I downloaded CUDA (Version 10.2) and CuDNN, I have moved files from CuDNN library to CUDA folders and added following variables to PATH:

C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2\bin
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2\libnvvp

After running following code in anaconda, I get following response:

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

I am guessing the main problem is "Could not load dynamic library 'cudart64_101.dll'; dlerror: cudart64_101.dll not found" , so I went to my CUDA bin folder and there is no cudart64_101.dll , but there is cudart64_102.dll . What should I do to fix this?

Answer 1

The last three numbers in your .dll files should point you to a solution. TensorFlow is looking for CUDA 10.1, but you have installed CUDA 10.2.

Re-installing CUDA and the compatible cuDNN should fix your problem. See the TensorFlow installation tutorial for the exact versions of CUDA and cuDNN. You'll probably have to reinstall TF as well, after installing the correct CUDA libraries.

Hope this helps.