Depends on how you define “accurate”. Even full ray tracing is just an approximation based on relatively few light rays (on an order of magnitude that doesn’t even begin to approach reality) that is deemed to be close enough where increasing the simulation complexity doesn’t meaningfully improve visual fidelity, interpolated and passed through a denoising algorithm. You can do close enough with a clever application of light probes, screenspace effects, or using a second camera to render the scene onto a surface (at an appropriate resolution).
That’s true, but after a few frames RT (especially with nvidia’s ray reconstruction) will usually converge to ‘visually indistinguishable from reference’ while light probes and such will really never converge. I think that’s a pretty significant difference.
Reflection probes are one way. Basically a camera drawing a simpler version of the scene from a point into a cubemap. Decent for oddly shaped objects, although if you want a lot of them then you’d bake them and lose any real time changes. A common optimisation is to update them less than once a frame.
If you have one big flat plane like the sea, you can draw the world from underneath and just use that. GTA V does that (like ten years ago without RT), along with the mirrors inside. You could make that look better by rendering them in higher resolution.
Where RT is visibly better is with large odd shaped objects, or enormous amounts of them. I can’t say it’s worth the framerate hit if it takes you below 60fps though.
I haven’t personally played a game that uses more than one dynamic reflection probe at a time. They are pretty expensive, especially if you want them to look high resolution and want the shading in them to look accurate.
What I’m talking about is drawing accurate reflections and I don’t know any other technique that produces the same accuracy as RT
Depends on how you define “accurate”. Even full ray tracing is just an approximation based on relatively few light rays (on an order of magnitude that doesn’t even begin to approach reality) that is deemed to be close enough where increasing the simulation complexity doesn’t meaningfully improve visual fidelity, interpolated and passed through a denoising algorithm. You can do close enough with a clever application of light probes, screenspace effects, or using a second camera to render the scene onto a surface (at an appropriate resolution).
That’s true, but after a few frames RT (especially with nvidia’s ray reconstruction) will usually converge to ‘visually indistinguishable from reference’ while light probes and such will really never converge. I think that’s a pretty significant difference.
Reflection probes are one way. Basically a camera drawing a simpler version of the scene from a point into a cubemap. Decent for oddly shaped objects, although if you want a lot of them then you’d bake them and lose any real time changes. A common optimisation is to update them less than once a frame.
If you have one big flat plane like the sea, you can draw the world from underneath and just use that. GTA V does that (like ten years ago without RT), along with the mirrors inside. You could make that look better by rendering them in higher resolution.
https://www.adriancourreges.com/blog/2015/11/02/gta-v-graphics-study-part-2/
Where RT is visibly better is with large odd shaped objects, or enormous amounts of them. I can’t say it’s worth the framerate hit if it takes you below 60fps though.
I haven’t personally played a game that uses more than one dynamic reflection probe at a time. They are pretty expensive, especially if you want them to look high resolution and want the shading in them to look accurate.