Is it possible to force your game to render supersampled? IE: have it render the image at 2x or 4x the actual resolution of your display, then downsample to the display resolution? FXAA is bad looking in general, TemporalAA is terrible on anything that isn’t a perfectly static scene. I cant figure out a way to do MSAA, so… Supersampling? Or if someone can tell me how to get MSAA working Id also be very grateful.
Thanks for any help you guys can give.
Afaik, you can’t since the engine is deferred. Well, with a little bit of work, it is possible to implement it but it’s not worth and deferred engines are not really made for that. Epic are really loving their Temporal AA and FXAA is there as a replacement for low end hardware so I don’t see a future for MSAA.
I can’t guarantee my answer is correct through
I’ll be honest and say I don’t know why being deffered would prevent a view being rendered at a higher res and then downscaled. Nor do I undertsnad my MSAA is not possible in Deffered, but I’ve encountered that in my use of Unity in the past. You are probably correct, but would you mind explaining?
MSAA works naturally in forward rendering because it was meant to work this way and is implement in the rasterizer. In deferred rendering, you kinda “bend” the pipeline by doing something that is not supported, that is, all shading is done using parts of a GBuffer. If you use MSAA on these buffers, the data they hold won’t be valid anymore because it has been smoothed out. For MSAA to work, you’d need to have buffers alot bigger than that, hitting the bandwidth and the GPU memory hard, really hard. There is a possiblity to manually resolve MSAA but it is not really convenient and probably unsupported on some platforms. Also, Temporal AA allows for specular antialiasing (I believe), while MSAA does not.
Again, I’m not an expert in this field so I can’t guarantee all of this is 100% true
I’m no expert on this subject either, but supersampling is way to expensive to be of any practical use in most cases. Technically, deferred rendering doesn’t stop you from rendering at 4x resolution and downsampling, so yeah.
Hardware MSAA is not possible with deferred rendering, since lighting is calculated on the G-buffer after the buffer is resolved (downsampled).
However, for tile-based deferred shading using compute shaders, there’s a MSAA technique that determines the edges of surfaces using ddx and ddy (link). Those pixels are then packed and multi-sampled individually. The results look good, but I’m not sure how well that method scales with modern graphics hardware and if it would work with the UE4 rendering pipeline. The same method was used for Frostbite/BF3.
Gutted… FXAA is average at best and temporal looks nice for some things but anything like additive rendering just looks terrible when it moves
Actually supersampling is supported by default. Simply raise screen percentage above 100 in your post process settings.
Sure it can get heavy when you raise it too much, but when combined with the post AA methods, just setting it a little above 100 can help reduce the disadvantages of those techniques.