Azdul
Magister
There are two parts to raytracing - tracing real rays, and extrapolating the results to simulate much larger number of rays.The way RTX does raytracing is kinda retarded to be honest. Next gen raytracing is going to be much more efficient than that. It will also utilize cpu cores and won't run on gpus alone. Let us wait for the hardware to arrive at our hands before we jump to conclusions.
The first part is embarrassingly parallel algorithm, and can be run on either GPU or CPU. In both cases it is equally hard, as it need constant access to complete geometry of the scene - it needs access to large memory structures than do not fit in low level cache. Intel used raytracing to demonstrate advantages of Larrabee architecture, because it was particularly good fit.
If you use anything else than 512 Larrabee cores used for Intel demos, you are limited in the number of rays you can reasonably trace. That's where second part comes in - extrapolating the results, and tensor cores of RTX are reasonably good at it.
What is possible without hundreds of cores and without RTX ? Probably reflections not limited to screen space on some surfaces, as seen in PS5 gameplay footage and AMD demos, and not much more.