New D3D FSAA Viewer

[MuFu]

Temporal AA can be enabled on R3x0 using the currrent drivers.

I'm going to check this out! Hard to believe it works correctly with R3x0 - why wouldn't it though? Hmm...

 

[KimB]

Why would you want the sample pattern to change frame to frame? Wouldn't that create extra temporal aliasing? Fractional pixel popping when the camera is still?

If the framerate's high enough it couldn't be noticeable. Remember that the "flicker frequency" is all about detecting very significant differences in brightness (i.e. 60Hz causes noticeable flickering on a monitor because it goes from bright to almost completely black: on darker images the flickering isn't nearly as noticeable).

I doubt that you'd need framerates higher than 30fps or so for the flickering of a frame-to-frame changing pattern to become unnoticeable for all but the highest-contrast situations.

 

[JCLW]

Don't LCD displays flicker pixels rapidly in order to display their 16.7 million colours? Does anybody know how fast they flicker?

 

[KimB]

Don't LCD displays flicker pixels rapidly in order to display their 16.7 million colours? Does anybody know how fast they flicker?

I really doubt it. Consider: old LCD displays can take a large portion of a second for the color of a pixel to change. If they used a "flicker" to achieve the different colors, they could change essentially instantly.

I believe what you're seeing is a surface that has an opacity that is voltage-dependent.

 

[MuFu]

Temporal AA can be enabled on R3x0 using the currrent drivers.

I'm going to check this out! Hard to believe it works correctly with R3x0 - why wouldn't it though? Hmm...

Well FSAAViewer shows the sample positions switching per frame at least. Blimey.

 

[Geeforcer]

Mufu, this is 6x, right? The pattern looks strange.

 

[MuFu]

Yep, 6x. Looks like completely new sampling patterns.

 

[3dilettante]

That's pretty nifty.

Just to recap, doesn't the R3xx permit the programmer to choose from various sampling patterns? Perhaps the driver just emulates the temporal method by cycling through the various patterns.

Perhaps the reason why it's not an official option is that there is a measurable amount of overhead associated with switching patterns on the fly. Are there other reasons why this wouldn't be implemented? Perhaps R420 might possibly eliminate this stumbling block?

Any performance data on using this method as opposed to a conventional one?

 

[MuFu]

HOLY CRAP, IT WORKS IN UT2004!

Absolutely positive - went from 4xT to 2x and the difference is so obvious. Plus you can see the tiniest bit of shimmer sometimes on high contrast edges (e.g. lanterns in dm-rankin) but it's not noticeable when moving or on low contrast boundaries.

HOT DAMN!

 

[Althornin]

why dont you post how to enable it?

 

[MuFu]

Because I thought I'd better check first whether it's ok to post.

Going to double check now because I still can't quite believe it. :?

 

[JCLW]

Don't LCD displays flicker pixels rapidly in order to display their 16.7 million colours? Does anybody know how fast they flicker?

I really doubt it. Consider: old LCD displays can take a large portion of a second for the color of a pixel to change. If they used a "flicker" to achieve the different colors, they could change essentially instantly.

I believe what you're seeing is a surface that has an opacity that is voltage-dependent.

I hate to refer to Tom's Hardware of all places, but they claim that almost all LCD panels are really only 18bit (262,144 colours), and use dithering to display 16.2m colours. http://www4.tomshardware.com/display/20031105/lcd-06.html

 

[3dilettante]

I hope ATI can think of a way to lessen the flickering with high contrast if they ever bring temporal AA into official use. Perhaps a temporal buffer that stores gamma information?

 

[MuFu]

Yeah, no doubt about it - a degree of awesomness is occuring.

With 12xT, the flickering is barely noticeable - even on high contrast edges. By far the best illustration of it working is 2x vs 4xT.

 

[3dilettante]

Is there a significant performance delta between a temporal setting with X number of samples per frame and a standard pattern with same number of samples?

Sure sounds awesome. Makes me wish I had up to date hardware (more so than usual ).

 

[digitalwanderer]

Yeah, no doubt about it - a degree of awesomness is occuring.

With 12xT, the flickering is barely noticeable - even on high contrast edges. By far the best illustration of it working is 2x vs 4xT.

Dang it MuFu, share the wealth!!!!

 

[BRiT]

Yeah, no doubt about it - a degree of awesomness is occuring.

With 12xT, the flickering is barely noticeable - even on high contrast edges. By far the best illustration of it working is 2x vs 4xT.

So how do the other modes compare, such as 4xT & 4x, 8xT & 6x? Certainly some sweet looking FSAA is in store for us. 8)

 

[3dilettante]

Here's a question: how can the effectiveness of this method be shown via screenshot since it operates over multiple frames? Damn, you might need an animated jpeg or something with a bunch of frames looping around while standing still.

Interestingly, looping it and finding if there is a number of frames that makes the looping appear seamless might indicate if the card is cycling through all those predefined sampling patterns. Assuming it doesn't just pick through them at random.

edit: Crap, even if animated, how are you going to get it to display fast enough to even try to show it? A vid capture probably wouldn't help either. How would this ever be evaluated?

 

[Tim Murray]

He's working on it now. Damn, I want to see this. Exciting times.

 

[101]

Wow that's a nice idea. Temporal dithering of the edges with a small number of samples could provide a noticeable improvement. The flickering concerns me tho, especially at choppy 30-40 FPS type of rates.