Temporal AA with SSAA

[Killua]

2) can 2x temporal AA be programmed to use the regular 4x AA pattern ( so that the two pixels on scene 1 and 2 pixels on scene 2 = the 4x aa pattern - I am not sure if this is currently what happens....)?

that is whats happening for 2xAA T2

 

[bloodbob]

is there a reason they don't expose it (in windows)?



is there some registry hack I can do?!?

<emotional breakdown>
ALL I WANT IS MY SSAA!!! DAMN THE PERFORMANCE HIT!!!
</emotional breakdown>

1) Performance, useful only in some older games
2) No

Stop spreading ATIs FUD

It is useful generally anywhere their is a alpha test.

 

[Joe DeFuria]

It is useful generally anywhere their is a alpha test.

From a quality perspective, yes. From a performance perspective, no.

 

[flick556]

I would like to see the options enabled with messages about the performance cost.

Just imagine a temporal mode equivalent to NVIDIES 6x mode

4x MSAA + 2x SSAA + 2TAA = 12X Mixed AA mode

Thier are games where x800 get's 200+ fps we need options that will increase the quality for these situations.

 

[o.d.]

ATI has stated that they do not implement the said feature because not many people submit it in the Catalyst Crew form as a requested feature. So, imho, instead of petitioning for it someplace where it doesn't matter, petition for it where it does: in the Catalyst Crew page

catalyst crew page linky?



EDIT: found it here

 

[Colourless]

Supersampling using temporal AA might seem like a nice idea but you need to remember that Supersampling does more that just AA alpha test edges too. It will change the texture filtering across the entire screen since the entire scene is being shifted slightly every frame. This could introduce an excessive amount of flickering, which can be quite strange to look at.

 

[Simon F]

Supersampling using temporal AA might seem like a nice idea but you need to remember that Supersampling does more that just AA alpha test edges too. It will change the texture filtering across the entire screen since the entire scene is being shifted slightly every frame. This could introduce an excessive amount of flickering, which can be quite strange to look at.

Err why? Surely if that were occuruing then that would imply that your texturing filter system is unstable and hence itself generating aliasing.

 

[bloodbob]

It is useful generally anywhere their is a alpha test.

From a quality perspective, yes. From a performance perspective, no.

Can you tell me anywhere that AA speeds things up? what else would we be talkign about?

 

[Simon F]

It is useful generally anywhere their is a alpha test.

From a quality perspective, yes. From a performance perspective, no.

Can you tell me anywhere that AA speeds things up? what else would we be talkign about?

I know you are probably just refering to (super|multi)sampling AA, but MIP mapping of textures is one place where AA speeds things up enormously.

 

[Colourless]

I wasn't meaning to imply that aliasing is being generated.

A half a pixel shift is going make the bilinear filtered result be different. The sampling point in the texture has changed so what you will get will be different. May not be 'hugely' different but it is still different enough that high constrast edges in texture can appear to flicker ever so slightly. Higher refresh rate of course will lessen the impact.

 

[Xmas]

Seems like my post has been overlooked...

"Temporal Supersampling", i.e. jittering the texture sample position/geometry per frame, is damn simple to implement. I was wondering why they haven't implemented it at least as an option yet. Sure, it looks horrendous when the frame rate drops. But so does temporal AA anyway. One problem is that you cannot really adjust the texture LOD, because you'd get shimmering in motion. And if you keep the LOD at the same level, texture quality gets slightly worse, most apparent on text. But it can fix alpha test edges.

 

[LeGreg]

Err why? Surely if that were occuruing then that would imply that your texturing filter system is unstable and hence itself generating aliasing.

Some cases are more prone to big variations in the final color: point sampling or no mipmapping, alpha testing.

 

[GraphixViolence]

"Temporal Supersampling", i.e. jittering the texture sample position/geometry per frame, is damn simple to implement. I was wondering why they haven't implemented it at least as an option yet. Sure, it looks horrendous when the frame rate drops. But so does temporal AA anyway. One problem is that you cannot really adjust the texture LOD, because you'd get shimmering in motion. And if you keep the LOD at the same level, texture quality gets slightly worse, most apparent on text. But it can fix alpha test edges.

I can't imagine this would look very good with just a single sample per pixel unless the frame rate was very high. And it would also blur out edges you actually want, like fine text. It could work if used selectively by an app on only the polygons that need it, but it would be much tougher to make something like that work as a control panel option.

Temporal AA works better as the number of samples increases, so the differences from frame to frame are minimized. And I can't think of a common case where blurring the edges of polygons would be detrimental to image quality.

 

[Lecram25]

We'll be seeing some temporal AA with SSAA soon.

 

[Tim Murray]

We'll be seeing some temporal AA with SSAA soon.

"Volaaaaaaari..."

Isn't that super...subsampling?

 

[Simon F]

Err why? Surely if that were occuruing then that would imply that your texturing filter system is unstable and hence itself generating aliasing.

Some cases are more prone to big variations in the final color: point sampling or no mipmapping, alpha testing.

And how is averaging these over time fundamentally worse than the errors due to the lack of antialiasing that would otherwise be going on?

 

[Basic]

If you're "shaking" the edges, you get a flicker at the edge.
If you're "shaking" the texture, you get a flicker over the whole area.

The flicker from shifting a texture half a pixel could give a intensity shift on par with the "edge flicker", but if you get it over a larger area it's much more visible. Just look at FSAAviewer, and compare the flicker on the edge with the flicker on the dots in the sample pattern.

I'd say it's still worth trying though.

 

[o.d.]

I'd say it's still worth trying though.

thats the spirit!

just give it a shot and lets see what happens. If it looks like @ss (in your opinion) then don't use it.

 

[Lecram25]

We'll be seeing some temporal AA with SSAA soon.

"Volaaaaaaari..."

Isn't that super...subsampling?

no no, lol. More like on teh doodoo 5.

:shifts eyes

 

[Bambers]

Hey, who remembers what AA the R200 was supposed to have?

Now, would that have actually provided better IQ than today's version of Smoothvision?

Dunno. Don't think it was ever exposed (definitely wasn't in launch drivers--you might as well have slapped your monitor for AA)--maybe it was later, but I've never really been able to find out (God knows I'm not installing a R200). Now, theoretically, I guess it could provide better IQ--but, it never would in a real life situation because you'd be able to use higher levels of multisampling without the framerate drops that would kill IQ.

no it didn't really materialise.

a non ordered (but not pseudo random) pattern appeared in the 7206 drivers (3864 or something for xp) for 2x and 4x quality modes only which in every driver after that reverted to only showing up in d3d in games that didn't use fog.


I'm not sure why they were removed but I did notice that at least in opengl games in the 7206 with 2x quality AA it did produce some very odd screen tearing on my 8500.