Hanners at EB: "Temporal anti-aliasing on R3x0 investig

[digitalwanderer]

Hanners just put up an article exploring ATi's recently discovered new temporal AA method.

As usual with Hanners, it's a great read and he makes it all easy to understand...peruse it at your leisure over here at EB.

 

[MuFu]

Nice write-up. I'd recommend setting the threshold to 55-59 if you use an LCD though. Setting it to exactly 60 will cause problems for many.

MuFu.

 

[Evildeus]

It was interesting.

However i've got one thing that i would have wanted to see: the same test with v-sync disable for normal AA.

 

[banksie]

Thanks for that link digitalwanderer. Most interesting and explains what Catalyst Maker has been hinting about as to something big coming.

Better yet is the backwards comaptible nature of the feature for any R300 card. Just another reason why the design of this card is an excellent example of engineering done right.

 

[BRiT]

Banksie, this is not the "something big" that's coming...

 

[MuFu]

That's true. Nobody knows about TAFâ„¢ yet (Temperamental Anisotropic Filtering).

 

[Tim Murray]

That's true. Nobody knows about TAFâ„¢ yet (Temperamental Anisotropic Filtering).

Yes, powered by Sirius Cybernetics Corporation's "Genuine People Personalities." Sometimes, he'll give you point sampling if he's feeling down, but other times, he'll give you 64-tap trilinear anisotropic filtering with no performance hit. He's a bit manic depressive.

 

[digitalwanderer]

Yes, powered by Sirius Cybernetics Corporation's "Genuine People Personalities." Sometimes, he'll give you point sampling if he's feeling down, but other times, he'll give you 64-tap trilinear anisotropic filtering with no performance hit. He's a bit manic depressive.

"A brain the size of a small planet and what would you have me do with it? Solve Einstein's Unified Field Theory? End world hunger? No, I'll make FarCry look nice and pretty for you instead....

....it's so depressing."

 

[Althornin]

lol nice one digi

 

[3dilettante]

Maybe somebody could take a long(ish)-exposure photograph of a screen running the time-staggered sampling. At least for non-moving parts of a game it could approximate what the intended effect would be.

Though the reviewer would have to make a pretty nifty contraption to isolate the screen and get proper images. Then they'd have to comment on whether the photo is representative, which leads back to the initial problem.

 

[Pete]

Or you could just blend two screengrabs from successive frames without moving the viewpoint. Still wouldn't aptly demonstrate how it looks in motion, where it'll probably have the most trouble.

Mufu captured the effect well with his Bink videos. I think.

 

[rainz]

Really interresting to read.. but there's almost no performance gain and you need a fast monitor ( 100hz ).. well excepted if you play 640x480 :?

RainZ

 

[3dilettante]

Or you could just blend two screengrabs from successive frames without moving the viewpoint. Still wouldn't aptly demonstrate how it looks in motion, where it'll probably have the most trouble.

Mufu captured the effect well with his Bink videos. I think.

The shortcoming of screenshots and video clips is that they don't accurately simulate the effect TAA exploits.
A screenshot is static, there is not running together of frames by the eyes that is necessary to show what the intended effect is.

Bink videos are better at showing the effect, but there are issues with trying to get the refresh rates to sync up with the player, which can create a differing impression than what real time viewing would do.

A photograph with a slightly longer exposure time will blend the frames together as the method intends (particularly if one is curious about how it might look on a slower-responding LCD), if the exposure is not too long or too short. It won't help in motion, true, but it would give an intermediate representation between screenshots and videos.

 

[kemosabe]

Really interresting to read.. but there's almost no performance gain and you need a fast monitor ( 100hz ).. well excepted if you play 640x480 :?

RainZ

Not sure I understand everything about the technique, but (in appropriate situations, as mentioned) there seems to be no performance loss for a higher effective level of AA. Is that not the equivalent of a performance gain? :?

 

[Bouncing Zabaglione Bros.]

Really interresting to read.. but there's almost no performance gain and you need a fast monitor ( 100hz ).. well excepted if you play 640x480 :?

RainZ

Not sure I understand everything about the technique, but (in appropriate situations, as mentioned) there seems to be no performance loss for a higher effective level of AA. Is that not the equivalent of a performance gain? :?

Looking at the graphs, the implication is that you get effective 4xAA using 2x temporal AA for the performance hit of 2xAA, and so on all the way up the scale.

 

[MuFu]

Yeah, those Bink videos are ok for getting across the general idea but, as 3dilettante said, they won't sync up properly (even if you set the refresh rate exactly you'll still have problems because the buffer flips and screen draws won't be aligned).

It is possible to capture if you make an AVI with sound that ReClock can latch onto. Even then there's the odd glitch. I might e-mail the ReClock author and ask him if he knows why it won't work with Bink. If it did (can't see any reason why not as it uses DirectDraw), that'd be perfect.

MuFu.

 

[digitalwanderer]

I dunno, I kind of like the fact that it's damn near impossible to show it/screenshot it....it adds to the whole mystique/charm of it.

Screw arguing with nVidia enthusiasts about it, it's one of those "if I have to explain it to you, you just wouldn't understand" things as far as I'm concerned. I was over at nVnews yesterday seeing the reaction of them to it and it amused me to no small end to see all the "problems" they found with it and the ways they "proved" it wouldn't work or look very good even though none of 'em had seen it or any demonstration of it!

This is gonna be a really weird couple of months coming up.... 8)

 

[Pete]

Yes, it's not a question of performance loss/gain, it's a question of better AA at the same speed. X800P/XT may be fast enough to make it work in all current cases.

(Whoops, this post was meant to follow BZB's.)

 

[Killer-Kris]

Also works with cat 4.3s

Just wanted to mention that this also works with the cat 4.3s.

Overall I'm rather impressed. The effect appears to be working in the AA tester program, and qualitatively appears to be working great in games. I haven't managed to see the flickering in the game itself, but if you watch closely you can sometimes see it on loading screens.

Yay for effectively 12x AA!!!

Edit: I'm almost waiting for this to show up on the front page, I think it would certainly warrant that.

 

[Eronarn]

I'm totally gonna test this out. Very interesting.