Well we have DirectML which is Direct X ver of machine learning API. So yes we could very well get machine learning reconstruction on AMD.
The sooner more devs adopt ray tracing the quicker more optimizations and tricks will be figured out.
Which is precisely why the dreams of a 60fps next-gen are already dead in the water. As long as they can hit a solid 30fps on the new systems, they'll throw everything and the kitchen sink at those graphics.
given how well stuff like DLSS and other upscaling techniques work, in addition to variable rate shading, and dynamic resolution, I don't see why 60fps is off the cards. a 2060 can hit well above 60fps in Wolfenstein with upscaling and VRS, and next gen systems will surpass that
Next gen will be interesting. It's going to be the most decisive in terms of performance vs fidelity. It's gonna be all over the place. I don't expect to see a ton of +90 to 120fps games for xbox series x and ps5, but I imagine there will be a decent amount attempted at least. Cross gen ports for the next 2-3 years could be the perfect example, like cod.
I know that, they can use it to generate probes more accurately. Basically my original post was talking about the wait times for both fully baked and probe methods and how RT is just going to cut that.
It really depends how it is used.
It's not just a form of reconstruction, it's an antialiasing solution too. I've said this to you multiple times already but you continue to spread BS.
To get anything based on ML from AMD we need to get AMD h/w capable of good ML performance first, APIs are secondary.
In many techniques AA and reconstruction or upscaling is combined. (Temporal injection, UE4 temporal upsampling etc.)
Its not BS. Nothing about DLSS is at all similar to supersampling
"Combined" doesn't mean that it's the same method as is the case with DLSS.
It is learning on supersampled images and tries to reconstruct what is learned from them. In case this wasn't clear from the technology description.
Should be possible.
I could imagine an AI could leverage a mix of screen space and world space hints to perhaps inference what a second or approximate third bounce of diffuse GI might look like
Specular bounce would be amazing as well.
Not really.
Looks like Ninja Theory is using ray tracing extensively in that Mara game and it shows. Basically photorealistic as a result of natural lighting and the added reflections. It seems like quite a leap.
take in mind, this is very early research and none of this will be used, especially not for real time rendering for video games in the coming 10 years.
I think it will do fine on next gen consoles. Whatever we have on PC right now shows that if anything even the smallest amounts of RT in a scene creates a "next gen" visual leap and going any further on the same implementation usually doesn't add much on top of that. So even with "weak" RT h/w consoles should be able to produce "next gen" visuals.
