A question to people who think RT will be the next big thing next gen: What about Tessellation?
Subdivision surfaces and geo displacement mapping has been a key feature of CGI way before ray-tracing was realistic even for high-end offline productions. And as such, it has always been taunted for the past 20 years as something that would make it's way into real time rendering, but never seemed to manage the transition fully. I mean, the first claim that Subdivision surfaces were the next big thing for real time rendering dates back to before 1999's Quake 3 for god's sake. We've been flerting with that idea ever since with every new console gen. PS2's VPUs were said to be able to do it, but rarely were actually given that task in released games (SSX resolved bezier surfaces in real time I think). Matrox Cards created their HW acceleration for them, later ATI with TruForm (Which was supported by a dozen or so PC games!) X360 did have HW tessellation in it which rarely got used, DX11 incorporated it in the standard specification, and even then, with both current gen consoles, PS4 and XB1, having HW accelleration compliant with DX11 standards, after 15 years of talk about this feature, most devs still show little interest in making that a foundational feature of their engines, even though that has been a bare minimum standard of CGI industry since it's beginning.
I know these two features: dynamic Tessellation (including subdiv and displacement) and Ray Tracing, are not completely comparable, and I can think myself of reasons why RT may be easier to adopt than the former, but it's worth remembering that even when a feature is highly desirable and it has been introduced in an actual product and implemented in actual games, it does not mean it will become an industry standard that fast. Ironically, the two things thing I hear from devs as reasons for Tessellation not having become completely ubiquitous still are that although reasonably programmable by now, the current standard is still not programmable enough for many desired applications, and the other one is that even being HW accelerated, it is bottlenecked by other aspects of the architecture in ways that make the cost not worth the end result. Both are very similar to complaints about the state of the HW RT implementation at the moment.
For next-gen, with all the talk about AMD's highly programmable next-gen geometry pipeline, and rumors of the architecture's impressive performance with micro-polygon rendering and massively complex meshes, it might be the case that Subdiv models will finally become the standard, and even then I'm a little skeptical. With that holy grail so eternally out of reach, It's hard not to be pessimistic about the other one that's arguably even holier, but also even further away.