A pity it's so expensive for the console because in the subpixel is really impressive... maybe on the ps3 they can tries something like custom MLAA of sony, spare in the spe to save precious ms, but I doubt cryengine it's so 'elastic' for this.
Alternative AA methods and their comparison with traditional MSAA*
- Thread starter mitran
- Start date
A pity it's so expensive for the console because in the subpixel is really impressive... maybe on the ps3 they can tries something like custom MLAA of sony, spare in the spe to save precious ms, but I doubt cryengine it's so 'elastic' for this.
You know that 16x CSAA is more like 4x MSAA? 16x QCSAA covers 8 samples.
And this AA is not designed for current gen consoles, but for PCs and next gen and is fully compatible with deferred engines.
Its really amazing for its cost and still can be upgraded further or just devs can allow to supersample image a little.
Ps. I'm pretty sure that i was posting it in Alternative AA thread.
FXAA doesn't do anything on subpixel level, but it dims single pixel sized features a little thus reducing shimmering in image.
So this is open source, available for any commercial developer to use without having to worry about any legalities?
1,3 ms on 470 for 1080p, don't have number for 720p.
Yep, just like FXAA, its free for all devs.
Right, that's the complaint... you can't make a claim like that without qualifying it, ideally in the same sentence. You have to understand that people are particularly sensitive to these sorts of claims for AA techniques because none of them look even as good as MSAA (let alone SSAA), yet since the start they all claim "looks as good as 16x SSAA". That's just a lie in the general case, and the case that they do look comparable is simple and irrelevant.
But it does look better than MSAA. In edge smoothing is beyond 8x MSAA and in subpixel reconstruction is somewhere between MSAA 4x and 8x.
And it wasnt single sentence and it was said in context. There are more claims in this paper and on their site that denies that its like 16xSSAA.
We've just got release of DX 11 enabled game on deferred shading based engine that has 4x MSAA + FXAA support, but 4x MSAA cuts framerate in half.
But with SMAA You would get the same combination but in cost a little higher than FXAA high.
I think that SMAA is awesome research and should be implemented in every possible game as an option.
Last edited by a moderator:
MSAA is slipping towards useless because of the increasing prevalence of shader aliasing.
This is definitely what people seem to miss a lot. The problem is how to present a comparison of each type of AA in motion....
We need a standard test scene engineered for the very purpose of testing AA, with aspects to test long edges, short edges, temporal aliasing, shader aliasing, etc. It could be rendered with 64x SSAA as a quality standard and each AA type can be accurately measured against this reference in terms of degree of error. We could then produce statistical evaluations of each AA type and rate them according to a metric that factors in both average differences and range blah blah you know the score. 
I guess one obvious although boring test case would be a black and white checkered quad that's viewed from a free-floating camera so it turns in every way. Ideally there'd be a few prettier reference scenes for static screenies for internet perusal too, with foliage, high-contrast shaders (metal?), near and distant surfaces, powerlines, and whatever other test situations we can think of.
I guess one obvious although boring test case would be a black and white checkered quad that's viewed from a free-floating camera so it turns in every way. Ideally there'd be a few prettier reference scenes for static screenies for internet perusal too, with foliage, high-contrast shaders (metal?), near and distant surfaces, powerlines, and whatever other test situations we can think of.Is there anyone that can make a test scene with engine e.g. HUMUS??... maybe in webgl format so people can throw in some shaders and try their AA method right in the browser. I mean..there are standard tests for filtering...why not AA which IMHO is a much worse a problem.
What are you basing that on? It needs to be seen in motion to judge subpixel quality and from the demo they posted, it's not even in the same league as 4x MSAA yet.
On AMD hardware, yes. There's clearly a slow path being hit there. On NVIDIA the hit is around 25% or less IIRC - i.e. similar to forward rendering and quite acceptable.
Not really true; no one is saying that MSAA by itself is sufficient. What we're really saying when we say we MSAA is good/necessary is two-fold:
1) You *need* a jittered sampling pattern. Anything that doesn't do this is basically garbage going forward. If you try to reconstruct solely based on a uniform screen-space pattern it's gonna look like crap.
2) You need adaptive sub-sampling. Some pixels simply need more shading than others and if you're unwilling to do that and handle the irregularity that incurs, it'll continue to look like crap in motion.
MSAA as exposed in DX10+ is a good tool for both of these. Ideally shader aliasing you handle by using better techniques that don't alias, but even in cases where you can't MSAA provides tools to do adaptive super-sampling with nicely formatted grids and good sample compression. Hard to imagine this is ever not going to be a critical tool in any good AA method.
I'm just thinking of how various recent games have MSAA support but it is of limited value because it does nothing for nasty shader aliasing the games have. RAGE, Hard Reset, DIRT3, etc.
The more advanced solutions seem ignored due to the PS360 stranglehold in addition I'm sure to he usual neglect anti aliasing suffers. MLAA and FXAA are easy and free, and work on the consoles, so this is surely why they are popular even though they suck.
The more advanced solutions seem ignored due to the PS360 stranglehold in addition I'm sure to he usual neglect anti aliasing suffers. MLAA and FXAA are easy and free, and work on the consoles, so this is surely why they are popular even though they suck.
For sure and they're obviously better than nothing. The only thing I'm objecting to is the characterization by some that they represent a step forward; they are actually a step back in visual quality although obviously higher performance.
But like I said, "shader aliasing" is not something that is addressed via one big hammer. You eliminate it at the source, by not writing terms that alias.
Last edited by a moderator:
The biggest source for shader aliasing is the current normal map approximation we use. Mipmapping is basically a precalculated way of antialiasing texture data. However mipmap downsampling works only well for color data, since source texture color has approximately linear effect on pixel final color. But for normal vectors, the results are wrong, and the errors are shown especially on surfaces with heavy specular. You should accumulate/average the final pixel colors (with specular + diffuse applied) instead of the normal vector directions to produce a properly antialiased image.
LEAN mapping pretty much solves the normal map antialiasing problem. The technique produces results that are often indistinguishable from the ground truth. However since it requires some extra shader math and more storage, not many games are yet using it.
LEAN mapping pretty much solves the normal map antialiasing problem. The technique produces results that are often indistinguishable from the ground truth. However since it requires some extra shader math and more storage, not many games are yet using it.
Similar threads
