In simple terms, if you target 60FPS and FSR gets you 180FPS, you are getting up to 2/3rds of your GPU back to do RT.
Ampere on the other hand doesn't care in that respect as it can run both RT and Tensor in parallel with CUDA.
The reason I call RDNA2 a console chip is that AMD chose their route for both power and cost budgets associated with the console/mobile markets, while Nvidia went all out brute force. I'm looking forward to RDNA3, hopefully free of console constraints where they will hopefully move RT outside of the compute core.