What do you mean by paradigm shift in software (genuinely curious)? The largest hurdle with RT is it's computationally expensive. Yeah, there are software oriented methods where specialized RT hardware doesn't need to be used to make these computations bearable. I was under the impression most of these were.... faking it. It's possible I am wrong though. I won't pretend to understand the concept they're using with these alternative methods fully.
A shift from rasterization based real time graphics, to path tracing based real time graphics. Which would mean effectively starting from scratch in terms of optimal core design. Throwing away all that is known about how to do computer graphics and starting afresh. Even the current RT cores still use some rasterization hacks to smooth things out.
It's the complexity of the various hacks that make rasterization computationally cheap, that make AAA-tier computer graphics increasingly expensive.
Path tracing simplifies that pipeline, which is why pre-rendered CGI-industry has moved from rasterization to path tracing years back, completely as far as I know. Not an expert, mind.
Such a shift would require a company with the necessary tech and the daring to do so. Problem is, that while nVidia has the tech, they don't have the daring. They like being on top too much to chance anything on a Hail Mary.
AMD, meanwhile, seems to be more willing to pull off risky moves. Not that they have a leadership position to lose. But while they may have the daring, they don't have the tech.
Well, as far as we know...
Well, I've often wondered if it wouldn't be in the best interest of AMD to go the other way with it.
[...]
Basically, swoop in under the radar and steal the rasterization crown for a lower price. Pursue the RT functionality as a side project until the hardware "catches up" to the intense computational requirements of RT.
That would be essentially like killing the golden goose. Exchanging the future viability of your company - or atleast the GPU side of it - for short term profits.
Besides, if AMD wants to win big, it needs to do the exact opposite. Ideal for them, would be to come out with a core design that can go full path tracing, but can also brute force emulate rasterization. A "miracle" core that can compete with or at least lose only slighty to nVidia Cuda and RT cores simultaneously. Thus is capable of ushering in the future.
I get the feeling that is AMDs strategy. To come up with a instruction set that allows them to do atleast a passable job of backwards compatibility, while possibly stealing the future crown from nVidia.
But that's for the generation after the next, unless AMD has managed to square that particular circle already. Probably not, but a man can dream.