Currently the Ambient Occlusion we have uses its "Low" sample count preset, as defined in the CameraPostprocessingManager.cs file.
Similarly to the FXAA changes I made a little while ago, this is just increasing a pre-existing quality setting, and is very unlikely to cause any breakage.
Comparison screenshots (all in the cloud home for the sake of consistency)
No AO
AO gets completely disabled when the intensity is set to 0, so this would be unchanged regardless of the sample count preset
100% AO intensity, current "Low" sample count
What the same scene looks like if AO is set to 100% in the settings.
100% AO intensity, "VeryHigh" sample count
Some areas are noticeably darker in this screenshot, due to the higher sample count causing more occlusion
150% AO intensity, "Low" sample count
This is using overdriven settings to try and reach a similar look to the "100% very high sample" with the current settings. It's not possible to get a 1:1 match between the settings though, and this method has its own issues; Some areas are incorrectly over-occluded, like the bush on the right having a small blob of ao that isn't in the "100% very high sample" screenshot. It's more noticeable at higher overdriven values like 166%, but I was trying to find the best balance of "close to the VeryHigh" setting and "Not having a lot of bad looking over-occlusion" It's also a bit more noticeable in game with the bush moving around.
150% AO intensity, "VeryHigh" sample count
This just looks horrible, and I'm including it both because I think it's funny to look at, and to also be an example of how anyone using overdriven AO settings would likely have to turn down said overdriven settings.
If it probably won't break, why make an issue instead of a PR like what happened with the FXAA/SMAA changes?
While raising this to the maximum "VeryHigh" setting is probably safe to do from a content breakage standpoint, I'm more worried about the potential performance hits. I made PRs for the FXAA and SMAA quality increases without a long discussion on performance because both of those options are pretty lightweight to begin with. It would be unlikely that increasing their quality would be enough to push a lot of people over the threshold of "consistent fps" to "noticeably choppy."
(like the only scenario where I was able to get FXAA at its absolute highest quality settings to cause any noticeable impact to performance was on my low end laptops in the local void world.)
Ambient Occlusion on the other hand, is a bit heavier than FXAA or SMAA. Granted, it's still not that heavy across the 3 machines I've tested, and the higher sample count only seemed to cause a minor performance loss, but it's still enough for me to be a bit more cautious with making this change. I'm also more cautious because I'm not sure if my usual "load up the cloud home" thing is the best method for testing how heavily AO effects framerate, so I have a couple questions about this idea.
1. Are there any particular worlds where AO is noticeably heavier compared to something like the Cloud Home?
This would be useful for testing how much the "VeryHigh" sample count actually hurts performance.
2. Could it be possible to have the sample count be a configurable setting in game, like Anti-Aliasing?
A sample count / quality setting is something common in many AO implementations, like Renderide's GTAO or various reshade AO effects, so in theory porting this kind of functionality shouldn't be too difficult. Even with that in mind, I understand if adding a new feature to be carried over to a new renderer is a blocker for this specific idea.
I currently have a branch on my fork of the renderer with the sample count set to VeryHigh, if anyone feels like building it for testing purposes.
Currently the Ambient Occlusion we have uses its "Low" sample count preset, as defined in the
CameraPostprocessingManager.csfile.Similarly to the FXAA changes I made a little while ago, this is just increasing a pre-existing quality setting, and is very unlikely to cause any breakage.
Comparison screenshots (all in the cloud home for the sake of consistency)
No AO
AO gets completely disabled when the intensity is set to 0, so this would be unchanged regardless of the sample count preset
100% AO intensity, current "Low" sample count
What the same scene looks like if AO is set to 100% in the settings.
100% AO intensity, "VeryHigh" sample count
Some areas are noticeably darker in this screenshot, due to the higher sample count causing more occlusion
150% AO intensity, "Low" sample count
This is using overdriven settings to try and reach a similar look to the "100% very high sample" with the current settings. It's not possible to get a 1:1 match between the settings though, and this method has its own issues; Some areas are incorrectly over-occluded, like the bush on the right having a small blob of ao that isn't in the "100% very high sample" screenshot. It's more noticeable at higher overdriven values like 166%, but I was trying to find the best balance of "close to the VeryHigh" setting and "Not having a lot of bad looking over-occlusion" It's also a bit more noticeable in game with the bush moving around.
150% AO intensity, "VeryHigh" sample count
This just looks horrible, and I'm including it both because I think it's funny to look at, and to also be an example of how anyone using overdriven AO settings would likely have to turn down said overdriven settings.
If it probably won't break, why make an issue instead of a PR like what happened with the FXAA/SMAA changes?
While raising this to the maximum "VeryHigh" setting is probably safe to do from a content breakage standpoint, I'm more worried about the potential performance hits. I made PRs for the FXAA and SMAA quality increases without a long discussion on performance because both of those options are pretty lightweight to begin with. It would be unlikely that increasing their quality would be enough to push a lot of people over the threshold of "consistent fps" to "noticeably choppy."
(like the only scenario where I was able to get FXAA at its absolute highest quality settings to cause any noticeable impact to performance was on my low end laptops in the local void world.)
Ambient Occlusion on the other hand, is a bit heavier than FXAA or SMAA. Granted, it's still not that heavy across the 3 machines I've tested, and the higher sample count only seemed to cause a minor performance loss, but it's still enough for me to be a bit more cautious with making this change. I'm also more cautious because I'm not sure if my usual "load up the cloud home" thing is the best method for testing how heavily AO effects framerate, so I have a couple questions about this idea.
1. Are there any particular worlds where AO is noticeably heavier compared to something like the Cloud Home?
This would be useful for testing how much the "VeryHigh" sample count actually hurts performance.
2. Could it be possible to have the sample count be a configurable setting in game, like Anti-Aliasing?
A sample count / quality setting is something common in many AO implementations, like Renderide's GTAO or various reshade AO effects, so in theory porting this kind of functionality shouldn't be too difficult. Even with that in mind, I understand if adding a new feature to be carried over to a new renderer is a blocker for this specific idea.
I currently have a branch on my fork of the renderer with the sample count set to VeryHigh, if anyone feels like building it for testing purposes.