In articles you did read. Did any of them actually cared to go into simplest of math? How many intersections/hit checks even RTX GPUs can do in reality vs marketing? How many rays you need per different type of ray use? (reflections/refractions/lighting/...) How does that change when you combine more of those technologies and are forced to actually add one more bounce. Realistically speaking, if game does not try to be DX-R showcase and uses minimal (real-world like) lighting for all objects. Then it stays on the ground with as few reflective objects as can since in real world we do not really have that many mostly glass and gloss on cars. And finally adds refractions where we have mostly glass and water. This game may run reasonably good at 1080p resolution on RTX 2080Ti. And it will have to be done very, very carefully. Instead you have game like Control, shown in another thread, that uses DX-R on something as flat as floor where planar reflection with proper shader can deliver exactly same result at fraction of performance cost. Even that SSR implementation they are giving in comparison could have looked same way as DX-R with proper shader... except left and right edge of screen.