Tensorflow 2, cuda, cudnn and tensorrt installation


I have a serious problem with all the versions and the non coherent installation procedures from different sources. It is not possible to find a solution to install tensorflow2 with tensorRT support.
I have a PC with:

4090 RTX

Linux aiadmin-System-Product-Name 6.5.0-21-generic #21~22.04.1-Ubuntu SMP PREEMPT_DYNAMIC Fri Feb 9 13:32:52 UTC 2 x86_64 x86_64 x86_64 GNU/Linux

nvidia-smi says cuda 12.2 but I didnt install it. It was installed with ubuntu since the system crashed if I did not allow drivers to be installed with it. Something about nouveau.nomodule=0, not working.

NVIDIA-SMI 535.161.07 Driver Version: 535.161.07 CUDA Version: 12.2

How do I proceed from here?
Which version of cuda, cudnn and tensorRT should I install to get it working with tensorflow2.
I want to run a training and detection with a resnet50v2 model.

Please write a guide or refer to a proper installation procedure. I have tried multiple ways and all fail.
I have anaconda installed so everything will be installed in a conda environment.

1 Like

This is have all users. If you install Cuda in base system then command nvcc -V (or nvcc --version) will say it

Some things you can watch in this table