I guest my best analogue is that we are in the N64, Playstation 1, and Saturn era of Ray tracing. Those were the first major 3D capable consoles, and they put out some great games, but I wouldn't say any of them did 3D really well. It wouldn't be until the Dreamcast, Gamecube, PS2, and Xbox that we got some decent enough looking 3D games.
I don't really know if this analog works that well, because with raster based 3D rendering, more power was meant to add more detail. Raster graphics basically throws away a ton of stuff before it actually draws the image. With more power, you can throw away less stuff. Using ray tracing, the amount of detail on the final image depends on how many rays you can shoot out, independent of how many assets there are in the scene or their quality. So basically, ray tracing gets "maximum quality" for free.
In any case, there's been some major improvements to optimizing the ray tracing pipeline. Like one of them being
ReSTIR.
However there's one other thing I want to bring into light. Raster based rendering is basically cheating on how to light things as much as possible. Sometimes those cheats are very effective. But there are some things in lighting that you can't really cheat your way out of, and at some point, you'll spend more processing power cheating than actually doing it the "correct" way.
I found something interesting while poking around. In NVIDIA's Rise of the Tomb Raider performance guide
in the section on dynamic shadows, they had this:
The guide was written in January 2016 with the GTX 980 Ti was one of the highest-end cards you could get. So by turning on dynamic shadows at all, you saw a hit of like 20-25% performance on medium, all the way up to 35% on the maximum setting. And unless you were using Very High, the quality of the shadows could turn ugly real fast if they had to project on a large area.
The way shadows are rendered here is they render the scene from the light's point of view and the shadow quality is bumping up the resolution of the render. This eats into both rendering time and VRAM. So if you want basically nice, clean shadows that look great no matter the distance, angle, or coverage, you have to spend a huge amount of resources to do so. And then do this for
every dynamic light source.
I had a chance to play around with a special build of UE5 that let you play around with ray tracing options and found what one of raster rendering's biggest weakness was. If you have something like 50 dynamic lights (which isn't hard to get to in gritty, urban settings), if you all want them to cast a dynamic shadow regardless of distance, then ray tracing starts beating raster rendering in performance. The raster rendering side has to re-render the scene (albeit, not entirely) 50+ times over and "very high quality" shadow maps can be 4096x4096.
You ever notice in some games that have large open spaces but have things that cast shadows on the ground (like say tall buildings or a cliff), and there's a glaring cut-off point where the shadows suddenly look uglier? This is why. The developer had to adjust the dynamic lights to make sure performance wouldn't get too bad.