I've posted on this before, but will try again. In rf1, it was impossible to get a fully anti-aliased image without using the method called "Super-Sampling. Anyone who has used this knows it will give you at least half the frame rate of regular Multi-sample AA. The effect you get without SSAA is that once the image is moving, you get shimmering at certain line edges that is not only distracting, but can make the entire game all of a sudden look a bit... butt ugly. The truly annoying thing about this is, it's always just a few specific entities in each track that makes the entire game look a garbled mess, with the most common culprit being the white lines on the roads, but sometimes bits of fencing, and other lines. I'm sure it is when a certain type of entity is used in the building tools, it's void of usual AA techniques, but i really think it needs to be resolved in RF2. Here's one example at Lime Rock: ---------------------- NO AA ------------------------ ---------------------- 4x AA ------------------------- ---------------------- 4x SSAA ---------------------- What you can clearly point out, is the orangey track line with no AA at all, and when 4x regular AA is applied, it gets no treatment whatsoever, none, nada, zilch. It happens on several different lines on the track, this being the easiest one to capture. Under this you see 4x Super-Sampled AA, where the offending line finally is treated. So I was wondering if the graphics guru's at ISI would comment on this problem dating back to rfactor 1, and why it's looking like it's going to hit rf2, too. The big problem is that 1-2 years from now no one will care as we will all be able to use Super-Sampling and get 100fps, but factor in super high poly mods a year from now, triple screens, those who will play with a 120hz monitor, and super sampling is going to be tough for a long time for a lot of users. Would really love to hear about this subject, as: 1) it is not just these singular lines on tracks that are affected, it is several different lines and different objects that amount to a lot of aliasing and shimmering, largely ruining the visuals. 2) rf2 looks twice as good if you enable super sampling, and you get a lot users who fire up rf2 for 15 seconds and announce to the world the graphics are poor, when this is likely attributing towards these statements. 3) performance issues where if i turn on SS now with my triples, i'll get under 25fps! I've uploaded a .psd with all three full images inside that you can switch between if anyone or the devs want to look at. This is with an AMD card (6970). http://www.mediafire.com/?84rxbqgtfccbsn0 Rfactor 2 looks gorgeous with 4xSSAA on, but i'm afriad i won't being enjoying visuals like this for years to come on triples, which sadly means rf1 currently looks better on my rig and will do for quite some time.
The simple answer is that standard or multi-sample AA only smooths polygon edges whereas super-sample AA also smooths texture edges such as those in textures that have transparency.
I did say it was the simple answer, there's a lot more to it than just that though, including how the lines have been mapped to their polygons, mip map bias settings on the materials etc.
yeah, if I recall the adaptive only works with textures like wire meshing and the aliasing is applied to the alpha channel mapping to smooth the borders of transparency?
I don't really know much about AMD's adaptive anti-aliasing other than I've never really found it to work very well. I just use super-sampling and suck up the frame rate hit... once you've seen 8xSSAA anything else looks rubbish!
How big of a difference in fps is there with 4xAA and 4xSSAA? http://forums.anandtech.com/showthread.php?t=262275 http://hardforum.com/showthread.php?t=1612937 Nvidia has supersample in AA transparency.
i'm a bit rusty on how they work, but MSAA takes away the aliasing for the actual geometry and not on colour values, while SSAA works on all pixels in a scene regardless of geometry, so all textures are treated too. By geometry i mean the wireframe meshes, like the car shape, etc. Adapative AA (transparent AA), works on transparent textures with an alpha channel on them, so things like 2D trees or wire mesh fences, etc. So that is why SSAA is so expensive - it has to do the calculations for many, may more pixels than MSAA. Earlier on a single screen, frame rate went from 95fps with MSAA to under 50 with SSAA Is there no other way for rf2 to bypass the AA limits on these various lines and use really thin polys or something? It really does spoil the visuals big time.
For me the only way i've ever been able to satisfactorily get rid of jaggies in ISI-based titles is to use the crazy 'Sparse-Grid Super Sampling' option. I'm not sure what this does (over and above normal super sampling) but it makes a big difference for me. I turn this on using Nvidia Inspector - i'm not sure if ATI/AMD cards have a similar setting or what it might be called. Using for example 2x or 4x Sparse Grid Supersampling in conjuction with 2x or 4x normal AA it is possible to still have acceptable framerates (above 60fps) but you really need a very powerful card (670 or 680 for Nvidia). Just thought I would mention the setting in case anyone doesn't know about it.
I think sparse grid would refer to the sampling position of the surrounding pixels. If you imigine all pixels are just colours, with an anti aliased jagged line just the old one with more softer colour gradients to smooth out the line. In this case it will use a different sample pattern around the pixel it needs to recolour to determine the new colour value. Not sure if that made any sense without a pic to go with it though