Too new driver?

Some time ago I trained this GAN and it worked fine, but now the IT department has upgraded the Nvidia drivers and CUDA version. The code still runs but the generated images are just noise. I cannot change the driver version, but I tried to install CUDA 10.1 (in addition to CUDA 11.2) and point PATH and LD_LIBRARY_PATH to the old CUDA, but still the generated images are just noise. Is it impossible to run code for an old tensorflow version (1.14) with recent Nvidia drivers (470.94) ? Or what do I need to do to make it work? I’m using anaconda and inside the environment I have CUDA toolkit 10.1, tensorflow 1.14 and CUDNN 7.6.4. Would it make any difference to use a Docker container?

I’m using a 3090 card, maybe that is the reason? Maybe it will work with 2080 or Tesla V100.

Yes it works on a 2080 Ti card…