Type of AA most beneficial for 360/PS3

[Butta]

It seems to me that AA causes the most problems in terms of bandwidth/tiling usage in next gen consoles (or at least it separates console GPUs from their PC counterparts). What is the most beneficial type of AA that can be used for either console (I often hear PS3 uses MSAA but fail to see how this is different from AA)? or is it better to just not use any AA (ala GeOW). Also are there any new techniques for limiting bandwidth usage that can be applied to these new consoles that are not in use today?

 

[Shifty Geezer]

They AA used is invariably MSAA as other types (specifically supersampling - not sure what other full-screen solutions exist) are too danged demanding. AA adds to image quality, but depending on the title, you may want to not use AA and redirect the resources on other effects.

 

[Squeak]

Seeing as AA is apparently so expensive to do that devs. seem to prioritize it below every other IQ effect, I can't help but wonder why cheap, "good enough" AA effects aren't used more.
I talking about stuff like finding edges in post processing of the z-buffer or looking for high contrast areas in the framebuffer.
It wouldn't get every jaggy edge but it would be a lot better than nothing.

 

[Pozer]

Expect Quincunx AA to make a big comeback in '08.










jk.

 

[Butta]

Expect Quincunx AA to make a big comeback in '08.


Had to look it up... is Quincunx available as an option in modern GPU's?

Found this:
Quincunx AA is a special mode in GeForce3 where two samples are generated per fragment, but 5 samples are averaged together when filtering down, thus increasing the number of effective samples per pixel and giving higher quality than 2 sample AA with only slightly greater cost. The term “Quincunxâ€￾ comes from the 5-pattern on the side of a die.

 

[Deusp]

Expect Quincunx AA to make a big comeback in '08.










jk.

Actually, for PS3 devs, if they merely want 2xAA, Quincunx is a pretty usable form of AA.

 

[Butta]

What about TSAA or TMAA?

 

[AlNom]

I talking about stuff like finding edges in post processing of the z-buffer or looking for high contrast areas in the framebuffer.
It wouldn't get every jaggy edge but it would be a lot better than nothing.

Speaking of which... I was trying to find out on google if anyone figured out what sort of selective AA Epic used for Gears of War; it's some sort of flag for static and dynamic objects, but I wanted to know how this affects the framebuffer and its size so that it fit within the eDRAM.

Instead I found a paper "Adaptive Z-Buffer Based Selective Anti-Aliasing" You'd need a membership or perhaps your school library has bought access for you. Anyways, they use a high pass filter on the z-buffer to get the edges and then apply some AA.

 

[rounin]

I won't mind Quincunx AA. Is it almost free without much complication?

 

[LunchBox]

I would love to see at least 4x FSAA but one can only dream...

 

[Tap In]

Speaking of which... I was trying to find out on google if anyone figured out what sort of selective AA Epic used for Gears of War; it's some sort of flag for static and dynamic objects, but I wanted to know how this affects the framebuffer and its size so that it fit within the eDRAM.

Instead I found a paper "Adaptive Z-Buffer Based Selective Anti-Aliasing" You'd need a membership or perhaps your school library has bought access for you. Anyways, they use a high pass filter on the z-buffer to get the edges and then apply some AA.

good info, thanks for the find.

I knew when people said GoW has no AA, that they were smokin' crack.

 

[Shifty Geezer]

Instead I found a paper "Adaptive Z-Buffer Based Selective Anti-Aliasing" You'd need a membership or perhaps your school library has bought access for you. Anyways, they use a high pass filter on the z-buffer to get the edges and then apply some AA.

Why AA based on difference in distance? If you have a black object in front, and a black object behind, there won't be any AA.

 

[Butta]

Nice article on Quincunx AA:

http://www.hardocp.com/article.html?art=MTI5

Is it true that it has the same quality as 4xAA with half the memory requirement?

 

[Simon F]

Nice article on Quincunx AA:

http://www.hardocp.com/article.html?art=MTI5

Is it true that it has the same quality as 4xAA with half the memory requirement?

The question is, "Are you comparing 'like with like'?"

Quincunx, in its basic form, is a 2x (sparse) supersampling pattern with a bartlett (aka tent) lowpass filter. If you compare that with 4x sparse sampling with a non-box (e.g. bartlett) filter, then it definitely does not have the same quality.

It may, however, be comparable to 4x regular grid with a box filter.

 

[AlNom]

good info, thanks for the find.

I knew when people said GoW has no AA, that they were smokin' crack.

It's too bad I didn't find anything related to Gears though. The search continues.

Why AA based on difference in distance?

Hm... if you don't have access to the paper, I can fire it off to you by e-mail. But it seems they were thinking about selective AA along the lines of FAA (Parhelia) and they came up with a relatively easier method by using the z-buffer. (I don't want to speak for them since I confess I'm not that knowledgeable in the area to give a conclusive answer ).

If you have a black object in front, and a black object behind, there won't be any AA.

True, but are those edges really necessary to be anti-aliased? I think the gist of the method is selective AA on edges that would be most noticeable to the end-user. Taking a game like Doom 3 or F.E.A.R. for instance where you'll get black on black often, I wonder if you would really notice the lack of AA most of the time. Just MHO though.

 

[zed]

I knew when people said GoW has no AA, that they were smokin' crack.

so youre accusing cliffy B of smoking crack, as he has stated theres no AA in gears of war

the above paper i assume can equally apply to barbie horse adventures as gears of war, since i assume no specific titles are mentioned

 

[Todd33]

so youre accusing cliffy B of smoking crack, as he has stated theres no AA in gears of war

the above paper i assume can equally apply to barbie horse adventures as gears of war, since i assume no specific titles are mentioned

CliffyB: We’re going to have some terrific motion blur and depth of field, but there probably won’t be Full Scene Anti-Aliasing. We’ll do things that’ll give us a similar effect, but again we’re speaking on the engine side of things so it’s not really my department. I’m not a big fan of jaggies though, so we’ll do whatever we can to get rid of those.

http://www.gaming-age.com/cgi-bin/s...iff Bleszinski - An interview in the key of B

It's not very definitive, he not a programmer.

 

[AlNom]

so youre accusing cliffy B of smoking crack, as he has stated theres no AA in gears of war

Mmmkay seemed to notice it first in one of the 720p screenshots. He verified it with an upclose photograph and others I know who own the game can confirm it too. The AA (probably 2x) is applied to static environment objects whereas dynamic objects such as your character have no AA applied.

Other screenshots exhibit this although it's difficult to get the right angle to see it clearly.

the above paper i assume can equally apply to barbie horse adventures as gears of war, since i assume no specific titles are mentioned

The above paper has nothing to do with Gears afaik. I only mentioned it as part of the story as to how I came to find the paper. Sorry if you misunderstood.

 

[Mmmkay]

It's really hard to spot mainly because of the art direction and its use of low contrasting colours. There isn't any AA applied to geometry behind light blooms either.

 

[betan]

Speaking of which... I was trying to find out on google if anyone figured out what sort of selective AA Epic used for Gears of War; it's some sort of flag for static and dynamic objects, but I wanted to know how this affects the framebuffer and its size so that it fit within the eDRAM.

Instead I found a paper "Adaptive Z-Buffer Based Selective Anti-Aliasing" You'd need a membership or perhaps your school library has bought access for you. Anyways, they use a high pass filter on the z-buffer to get the edges and then apply some AA.

Thanks.
I am curious though, how are they using HPFed zbuffer for AA? Just use it as a blur (or some other LPF) mask?

I assume selective multisampling would be very costly so when people say "Selective AA" does it imply selective low pass filtering only?

Why AA based on difference in distance? If you have a black object in front, and a black object behind, there won't be any AA.

AA? You mean "there won't be any aliases"?
AA is where ever you want it to be .
I think z-buffer makes sense though.