I tried several combinations from this list to try and activate GPU acceleration for deep learning. However, nothing seems to work with tensorflow.
The CUDA installation works fine (by fine I mean running nvcc --version for example prints the correct output) and tensorflow is installed correctly, but I can't figure out a way to test and see what goes wrong between them.
I usually test my build like this: tf.config.list_physical_devices('GPU') and expect to see my graphics card listed there (or, at least, not an empty list). I have a NVIDIA RTX 3060 Ti card, so as far as compatibility with hardware goes, I don't see a probleme there.
I work with Windows 10, I know a possible alternative would be WSL2 with conda environments, but I want to exhaust my possibilities with Windows first.
Have any of you got any working configuration (tensorflow, CUDA and CuDNN versions)?
Configurations tried:
tensorflow 2.13.0 -- CUDA 11.8 -- CuDNN 8.6
tensorflow 2.12.0 -- CUDA 11.8 -- CuDNN 8.6
I also tried some variations with CUDA 12 and the latest CuDNN, but neither those seem to work.
One problem I read about was that compatibility with Microsoft Store installed Python was questionable, so I reinstalled my envirnoment with an official release of Python3.11, but to no avail.