The main intent for DXR as written by MS is to try to keep DXR as flexible as possible in alignment with the way that GPUS are moving into compute and further away from fixed function. For those reasons the DXR doesn't actually create a 'Ray tracing' object, they're actually generic objects and the can be used however desired, if you want to use it for AI vision, sound, graphics, etc. I think it's entirely possible in the future that the vendors identify a way for the developers to create their own intersection tests that would be as fast as fixed function. How that's accomplished is beyond me, but I suspect this is the evolutionary path.So ... based on that flow chart, do you think it would be possible to mix and match different types of intersection test? E.g. use voxel (terrain) and tri (characters)? Err ... put both in the same acceleration structure and enumerate either type?
tldr; yes. Yes with you own intersection tests you can support non triangle objects. But it will likely be very slow. Developers are also allowed to
I believe the vendors will figure out a way to make programmable intersection tests just as efficient as fixed function, but that may take some time to happen.
The feature set needs to be on both, it's unlikely they'll restrict it just 1. And knowing PS5 will have it, I can't seem them impeding progress for their Windows strategy. DXR is still very much tied to 12, and 12 to windows 10. It also makes streaming more appealling to players if the games they are streaming are ray traced - if you don't have the latest hardware streaming suddenly becomes very attractive. They should align with Sony on this and have RT as a baseline. I think this is a good business decision and Sony signalling their Ray Tracing was an effective way to say, hey 'we are doing this, so let's align'. Or you know, they've been talking about alignment of features in backroom talks that are confidential.Hmm RTX 2060 is 6.5 TF. Lockhart is rumoured to be ~ 4TF, which on the surface of it doesn't look great. Then again, would Nvidia even try selling RT to people who didn't mind gaming at 1080p/30 fps? There's probably a point where they want the minimum die area and the maximum incentive to go up to a higher margin product like the 2060.
I don't know nvidia well, but being first to feature sets has served them well many times over. They are making big bank in the AI industry, and have found ways to bring tensor technology to the consumer space. That is interesting. to say the least, I'm not sure if that put pressure on AMD. I think MS is happy nvidia has moved quickly because they need developers to start coding for RT today to be ready for the console space in 2 years time.That's my thought too.
As an aside, I wonder if Nvidia tried to use hardware RT acceleration to flog Turing to either Sony or MS? That might have put some pressure on AMD as might favourable moves to MS for an evolution of DX. *shrug*
Nvidia had to have been working on this since 2016, at least.
Hmm. With how we see MS doing their BC 2 generations back, I don't think any design would have caused them any serious issue; though it may grievance Sony a touch more. Both teams are extremely talented, and often I would look at 'feature' issues as more as business problem than they have been technical.'ve been wondering if the strain of building console processors has pushed Navi back. Even without RT, having to change the Navi design to allow for the possibly for compatibility with PS4 and X1 might have been more than AMD were comfortable with. With the the Pro, X1X, and now PS5 and Xbox next(s) they've had a lot of custom work on their hands that requires ongoing compatibility.