RTX 3070 tensorflow low inference performance

Hello, I have one RTX 3070 and i have installed CUDA 11.0 with cudnn 8.1. I also have python 3.8.9 and tensorflow 2.4.0 installed via pip. When I try to run inference, everithing seems to be okay, but when I compare inference performance with my other graphic card in other computer (gtx 1660 Ti), RTX 3070 is slower. Can I do anything to improve this performance?

Best regards.

RTX 3070 is Compute 8.6 which is not supported until Cuda 11.1.

And do you know if CUDA 11.1 is compatible with Tensorflow? Maybe building Tensorflow from source instead of pip.