Curious - modifying code to utilise more modern version of CUDA

[phil-fw60]

I am a little confused about the different versions of tensorflow. I have tacotron running OK on one machine and decided to install a more modern 11 gb nvidia card in a second PC and copy the install from the first - given all the dependencies are working!
I keep hitting problems with the GPU. The code is still finding the cuda 10 xxx.dll files but fails to use the GPU and falls back on the CPU. My card is a GTX3060 and I suspect it does not want to play with the older CUDA dlls. Is there a solution for this? Thanks