Last time I used it the step is just "Choose your engine", in the command line plaidml first detect the drivers, then you choose which driver you want to use, after that the system will run according to your preferences. (This is reference to use plaidml btw, for using amd outside plaidml I heard you use ROCm but it's locked to enterprise grade gpu afaik)

But yeah, cuda literally just install tensorflow cuda and you're good to go (I think, I never do it myself since I only have laptop version of 1050 and my family's PC use Radeon)...