ATI Temporal AA...

[nst6563]

Can anyone explain just exactly WHAT the temporal AA is/does...plain english (well, mostly plain anyway)?

I know that on my 9600xt, if I enable 2xaa in a game like Ravenshield...I notice a slower framerate (this is at 1280x1024, 16xAF, 2xAA, other settings on quality, game settings all on high)...however when I run the same settings only with 2t2xAA it runs about as smooth as it does when I don't run AA and the jaggies are DRASTICALLY reduced...

whatever it is/does...I think it kicks ass...at least on my system....

thanks for input.


:EDIT: Ok, I played some more and took some screenshots...and the 2xAA looks really close to the 2t2xAA...but the 2t2 is a LOT smoother in playing. I notice hardly ANY slowdown.

Here are my screenshots...tried to take them as close to the same place as possible.
They're all about 420k in size. 1280x1024. They look a little worse in the jpg's than they do in the .bmp I guess b/c the compression.

shot1 2xaa
shot1 2t2xAA
shot1 2t3xAA


shot2 2xaa
shot2 2t2xAA
shot2 2t3xAA

shot3 2xaa
shot3 2t2xAA
shot1 2t3xAA

:EDIT: Fixed the pics. I just found a 2t3 in the driver options...so I'll try those

RS must stay at 60fps or above b/c the game ALWAYS looks really smooth....although, I'd have to run FRAPS to be sure....

:EDIT: I just added some 2t3 screenshots for those who say that "tAA doesn't show up in screenshots". They are friggin beautiful. Much smoother lines than 2xAA and gameplay is about the same as 2t2. So whatever they're doing, I like it. Whether or not it's every other frame, it looks damn sweet -especially the 2t3...compared to the 2xaa and 2t2.

 

[Yonkec]

im not an expert in any of that stuff, but from what ive read/heard it does 2 seperate AA algorithims and alternates them every frame so you should get the best of both? but you have to have v-sync turned on for it to work, that and it only works above 60fps, so if you hit a big slowdown somewhere it turns itself off temporarily. this sorta makes it look like your using better AA than you really are. or so ive heard ;D

btw, the pics with temporal AA dont load

EDIT: heres a quote from elite bastards on it http://www.elitebastards.com/page.php?pageid=4533&head=1&comments=1

"One of the impressive (but little discussed) features of ATi's R3x0 architecture is that it is capable of programmable AA patterns. Whereas most graphics boards only have pre-programmed sampling patterns, theoretically a developer (or indeed ATi themselves) could change to a different AA pattern whenever they chose. This feature has never really been explored by either ATi or developers... Until now, where it has become the basis for a new (and thus far only available via the registry) anti-aliasing mode.

So, how does it work? Basically, instead of using a single, static sampling pattern when AA is enabled, temporal AA allows the board to select from one of several patterns, which it can then switch between on a per frame basis. In other words, every rendered frame is antialiased slightly differently, using a different pattern. If at this point you are thinking that changing the pattern every frame would end up as a horrible, flickery mess, you'd be right. Until you examine the principle a little more closely that is....

What you need to remember here is that the user is not seeing a handful of frames every second, in most cases on a modern GPU you would hope to see some way over 60 frames per second being rendered by the card. When you reach these kinds of performance level, the changes in sampling pattern are happening so quickly, and so frequently, that the human eyes cannot detect the difference between the changes at all. Therefore, the eye sees all of the different patterns used to antialias the image at the same time, thus giving the impression of a higher level of AA than is actually in effect.

The one big potential caveat here is that if frame rates drop too low, then the effect becomes visible to the human eye and flickering will occur. Therefore, this new technique is obviously not going to be suitable for all titles, the user will need to pick and choose carefully those which will give sufficient performance for the method to function as it should. A side-effect of this is that V-sync has to be enabled when temporal AA is used (this is forced on in ATi's drivers when it is in use), and also requires a high refresh rate on your monitor to really do it justice (in the order of 100Hzedit: seems to do the job nicely). "

 

[V0ltage]

yea it does look mad nice, but the second it drops below 60fps theres tons of flicker :\

 

[CleanSlate]

temporal aa supposedly doesn't show up on screen shots....................................

 

[nst6563]

temporal aa supposedly doesn't show up on screen shots....................................

That would explain the small jaggies or rough edges...but then again, the same thing shows up in normal 2xAA on my screenshots. But like I said before, it appears worse in the .jpg's I linked than the original .bmp

 

[tranCendenZ]

Temp AA won't show up in screenshots. Easiest way to explain it is that it does AA on every other frame, and your eyes/brain blend this together to look like a higher level of AA with the performance hit of a lower level of AA. This isn't exactly right but its the easiest non-tech explanation.

Only problem is that when the framerate drops below 60fps, which happens in most games, you will see shimmering/flickering because the frames aren't moving fast enough and your eyes/brain detects the differences between frames. TempAA also can give some people headaches due to its flickering nature at framerates under 60fps.

All in all, its an interesting gimmick, but good old regular AA is superior.

 

[nst6563]

Temp AA won't show up in screenshots.

All in all, its an interesting gimmick, but good old regular AA is superior.

for the first line, see my edit with the 2t3xAA shots. Compare those shots to the 2xAA and 2t2xAA shots. It's a lot smoother...and apparently it DOES show up in screenshots.

also, this "gimmick" you speak of. If it works, why call it a gimmick? a gimmick is:
2. gimmick - any clever (deceptive) maneuver; "he would stoop to any device to win a point"

so, if it's an AA method that produces measurable results then it would not be a gimmick. If it WERE a gimmick, then so would Nvidia's new lightmapping methods.

 

[tranCendenZ]

so, if it's an AA method that produces measurable results then it would not be a gimmick. If it WERE a gimmick, then so would Nvidia's new lightmapping methods.

It's a gimmick because under real world game conditions it usually provides an inferior picture to standard AA due to shimmering/flickering when games drop under 60fps, because its rare that games stay above 60fps all the time.

And if the game has so much framerate to spare that it won't drop under 60fps ever, you will get significantly more benefit from supersampling than from an emulated higher level of multisampling like TAA.

 

[nst6563]

It's a gimmick because under real world game conditions it usually provides an inferior picture due to flickering, because its rare that games stay above 60fps all the time.

funny...I haven't seen a single flicker yet. I've played Ravenshield for about 4 hours today, DF:BHD, BF: Vietnam, and MVP2004. None of them flickered , all played really well, and all had anti-aliasing 2t3 going. But I guess those aren't real world gameing conditions now were they? If what you're saying is true, then surely I would've seen "flickering" since I know my 9600xt doesn't keep above 60fps all the time. I would possibly also have a headache...which I don't. So either A, there's no flickering, or B, I was above 60fps all the time in all those games (highly unlikely).

So I suppose Nvidia's new lightmapping methods are also gimmicks b/c they provide better results than ATI's.

If you can provide useful links to this "flickering" and gimmicks then that's fine, otherwise stop threadcrapping. I asked for information on WHAT [U IT[/U] was/did, not your opinion of it.

 

[tranCendenZ]

If you can provide useful links to this "flickering" and gimmicks then that's fine, otherwise stop threadcrapping. I asked for information on WHAT [U IT[/U] was/did, not your opinion of it.

lol fine you want links because you dont believe me, have links:

http://www.elitebastards.com/page.php?pageid=4533&head=1&comments=1

"The one big potential caveat here is that if frame rates drop too low, then the effect becomes visible to the human eye and flickering will occur. Therefore, this new technique is obviously not going to be suitable for all titles, the user will need to pick and choose carefully those which will give sufficient performance for the method to function as it should. A side-effect of this is that V-sync has to be enabled when temporal AA is used (this is forced on in ATi's drivers when it is in use), and also requires a high refresh rate on your monitor to really do it justice (in the order of 100Hz seems to do the job nicely).

Of course, due to the aforementioned prerequisites (the need for V-Sync and high refresh rates), Temporal AA is never likely to become the primary method of anti-aliasing used in ATi's drivers, when exposed it will most likely become an additional setting. There will also no doubt be purists who will always prefer to use the standard AA modes due to its being guaranteed flicker-free and use of a superior sampling pattern. On top of that, there are always going to be titles that push a graphics card (or indeed CPU) to its limit, and as a result pull framerates down to the point where Temporal AA is unusable."

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

"The benefits of temporal AA are not very noticeable when you are moving around fast in a game. We would really prefer more physical samples of AA rather than this method of AA so that there would be no shimmering, and so that it wouldn’t depend on your framerate being above 60fps. Temporal AA only works with Direct3D for now."

I never make things up when I state something.

 

[nst6563]

sweet. That's what I was looking for. The refresh rate may explain why I didn't see any flickering. I play with the refresh rate at 120hz (Nokia 445pro, I'm able to select up to 185hz @ 1280x1024)


I take the threadcrapping statement back now.

 

[CleanSlate]

for the first line, see my edit with the 2t3xAA shots. Compare those shots to the 2xAA and 2t2xAA shots. It's a lot smoother...and apparently it DOES show up in screenshots.

also, this "gimmick" you speak of. If it works, why call it a gimmick? a gimmick is:
2. gimmick - any clever (deceptive) maneuver; "he would stoop to any device to win a point"

so, if it's an AA method that produces measurable results then it would not be a gimmick. If it WERE a gimmick, then so would Nvidia's new lightmapping methods.

It's not a gimmick.. lol, it's a simple situation.. come on.. and there are measurable results just go into a game that allows demos, record a demo and watch it with and without temp aa on.. gah ppl and their naysaying.. sad.

 

[Yonkec]

I was lead to believe it turns off if frames drop below 60fps for more than a second or two

 

[CleanSlate]

lol fine you want links because you dont believe me, have links:

http://www.elitebastards.com/page.php?pageid=4533&head=1&comments=1

"The one big potential caveat here is that if frame rates drop too low, then the effect becomes visible to the human eye and flickering will occur. Therefore, this new technique is obviously not going to be suitable for all titles, the user will need to pick and choose carefully those which will give sufficient performance for the method to function as it should. A side-effect of this is that V-sync has to be enabled when temporal AA is used (this is forced on in ATi's drivers when it is in use), and also requires a high refresh rate on your monitor to really do it justice (in the order of 100Hz seems to do the job nicely).

Of course, due to the aforementioned prerequisites (the need for V-Sync and high refresh rates), Temporal AA is never likely to become the primary method of anti-aliasing used in ATi's drivers, when exposed it will most likely become an additional setting. There will also no doubt be purists who will always prefer to use the standard AA modes due to its being guaranteed flicker-free and use of a superior sampling pattern. On top of that, there are always going to be titles that push a graphics card (or indeed CPU) to its limit, and as a result pull framerates down to the point where Temporal AA is unusable."

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

"The benefits of temporal AA are not very noticeable when you are moving around fast in a game. We would really prefer more physical samples of AA rather than this method of AA so that there would be no shimmering, and so that it wouldn’t depend on your framerate being above 60fps. Temporal AA only works with Direct3D for now."

I never make things up when I state something.

These are all things that were said when it first came out.. ATI was upfront about this stuff,, how would that make it a gimmick.. so what you have to have more than 60 fps? play within yoru compter's known ability to where you know it wont go under 60 fps.. not a difficult thing to do, I do it in cs.

~Adam

 

[tranCendenZ]

These are all things that were said when it first came out.. ATI was upfront about this stuff,, how would that make it a gimmick.. so what you have to have more than 60 fps? play within yoru compter's known ability to where you know it wont go under 60 fps.. not a difficult thing to do, I do it in cs.

I feel its a gimmick because I fail to see the usefulness in it, unless you have a really low powered card that can't handle 4xMSAA

If you use a threshold, there will be artifacting when the card switches from TAA to regular AA. For instance, if your threshold is 60fps and there is a part of the game that wavers between 55fps and 65fps, you will see artifacting while the card switches back and forth between AA modes while you go back and forth over the threshold (not uncommon with AI, special effects, etc)

If you don't use a threshold, you will simply see artifacting at low framerates.

On top of that, you need to use VSYNC. Using VSYNC means that the framerate will progressively cut itself in half it cannot handle the specified refresh timing. This means, for instance, if your refresh rate is 100hz, your possible framerates will be 100fps, 50fps, 25fps, 12.5fps, 6.25fps, etc. So, instead of enjoying a comfortable 45fps, you will be forced down to 25fps when vsync is enabled. This can actually lead to slower framerates than if you used a higher level of normal AA with vsync disabled.

In the end, aside from the user who just can't afford 4x AA, I really don't see the purpose of TAA. The fact that it must be disabled under 60fps or shimmer (and the artifacting associated with either) coupled with the fact that it needs VSYNC and the performance issues that causes makes it an unattractive AA mode to me. The Radeon's 4x and 6x modes are much better choices for gaming, or in the case of the 6800 4x and 8xS.

 

[CleanSlate]

If you play games or with settings that rip apart your card so bad it can't do 60 fps... upgrade . I play well at 30 fps but I play to the best of my ability when I dont get lag from loading 1600by1200 frames at 30 fps... really 60 fps is what you hsould be shooting for with settings anyhow, just gives you a better reason then "I need high enough frame rates to play good".

I was playing DOD at 1280by1024 with 2x taa and 2x af and it dipped below 40 fps most of the time on a certain map from 72 to 40 switching quickly, I noticed no "artifacting". So, I dont see an issue with it yet, I'll let you know if I see a problem, until then your theory is jsut a theory. I use TAA when ever possible, halving AA computations for double the AA.. mmm it's a good thing in my book, even if it does have semi-strict requirements.

~Adam

 

[tranCendenZ]

If you play games or with settings that rip apart your card so bad it can't do 60 fps... upgrade .

Ah, I guess you have not heard of the titles "Far Cry," "Doom3," and the upcoming "Half Life 2" then Even with my P4 3.2ghz cpu and 6800GT, I can only dream of having constant 60fps +aa/af at 1600x1200 in FarCry or Doom3. 1600x1200 with 2xAA/8xAF in FarCry is very playable and looks awesome, but no way in hell does it maintain 60fps - the average is probably more like 45fps. In fact sometimes it dips below 30fps. And thats on a fast setup.

I play well at 30 fps but I play to the best of my ability when I dont get lag from loading 1600by1200 frames at 30 fps... really 60 fps is what you hsould be shooting for with settings anyhow, just gives you a better reason then "I need high enough frame rates to play good".

The problem is with vsync on, at 120hz refresh rate, 58fps = 30fps. With today's games you have to crank down the detail and resolution to stay above 60fps all the time.

And, for instance, I'd much rather play FarCry at 1600x1200 2xAA/8xAF maximum detail at 30-45fps than FarCry at 1024x768 4xAA/8xAF with high detail at 60fps. The only time I've seen a resolution jump be defeated by antialiasing is when supersampling is used in close resolutions (ie 1280x1024 versus 1600x1200) on games that have a lot of foliage and aliased textures - supersampling at 1280x1024 can look superior to any level of multisampling at 1600x1200 in certain games - call of duty comes to mind. Of course this has nothing to do with TAA since TAA is multisampling so I am getting off track.

I was playing DOD at 1280by1024 with 2x taa and 2x af and it dipped below 40 fps most of the time on a certain map from 72 to 40 switching quickly, I noticed no "artifacting". So, I dont see an issue with it yet, I'll let you know if I see a problem, until then your theory is jsut a theory. I use TAA when ever possible, halving AA computations for double the AA.. mmm it's a good thing in my book, even if it does have semi-strict requirements.

well my "theory" is proved in the above articles. But you can do as you please of course

 

[nst6563]

I didn't mean to start an argument about what card is better and all this fps crap, I simply wanted to know what this stuff was b/c it seemed to do wonders on my system.

If the game looks good and is very playable...then I don't give a rats ass what fps I'm getting. To me, it's just like a Sandra or 3dmark score...it doesn't mean shit as long as you're happy with your systems perfomance.

 

[Smith]

The problem is with vsync on, at 120hz refresh rate, 58fps = 30fps. With today's games you have to crank down the detail and resolution to stay above 60fps all the time.

VSync doesn't lower your frames per second, it limits it. If you get 40 fps with vsync off, and then turn it on at 60hz, then it'll still be 40 fps.

If you get 70 fps with vsync off, and then turn it on at 60hz, then it'll be 60.

And if you always got like high 60's, or low 70's, and your refresh rate it 60hz, then most likely, it'll always stay at 60 fps.

 

[burningrave101]

VSync doesn't lower your frames per second, it limits it. If you get 40 fps with vsync off, and then turn it on at 60hz, then it'll still be 40 fps.

If you get 70 fps with vsync off, and then turn it on at 60hz, then it'll be 60.

And if you always got like high 60's, or low 70's, and your refresh rate it 60hz, then most likely, it'll always stay at 60 fps.

No. If you enable v-sync and your fps dips below the set refresh rate of your monitor, it will cut your fps in half.

V-sync is only useful if your playing an older game where you get really high fps or if you have an LCD. LCD's cause alot of tearing because their response time isn't very high in comparison to CRT's refresh rate.

It definitely looks much better. However, if you have a graphically intensive game that varies greatly in fps, it might appear that your display is stuttering somewhat. What I mean is, since I usually run at a refresh rate of 85hz, if the card can manage at least 85fps it'll lock it at that. However, if it were to dip down below 85, v-sync will lock it at 1/2 that, say 42 for sake of argument, to avoid that "tearing" effect. This leads to what I see as a stuttering effect. For this reason I usually use it only on older games.

True, this is what triple-buffering is for. Instead of waiting for the next retrace, thus incurring the delay, the app starts rendering to a 3rd buffer, and when the vertical retrace occurs the drive then automatically displays the 2nd buffer, while the app is still rendering to the 3rd. Its stops the frame-rate going 85 -> 42 -> 24 -> 12 etc.. etc..

Triple buffering helps to prevent the massive fps drop from enabling v-sync.

I dont use v-sync and i notice higher fps and no tearing effect. I find it useless for todays games on high-end CRT's.

 

[Smith]

No, I don't want to argue, but you're wrong.

Vsync doesn't cut your fps in half, it just keeps it from going over the moniter's limit. It doesn't slow you down at all.

 

[Koz]

I guess The Tech Report sees differently from you.

When you get into the game, the first thing you're going to want to do is enable vertical refresh sync, or vsync. Without vsync enabled (and it's disabled by default) you'll see lots of tearing (a sort of screen fragmentation) in DOOM 3. In my experience, setting vsync to "always on" in the ATI and NVIDIA drivers doesn't help. The game's default setting overrides the drivers, no matter what.

There still are tearing effects in games it just depends on how they've been coded it appears.

 

[TheFury]

luckily lcds aren't affected much by the TAA "flickering" at <60 fps

 

[nst6563]

luckily lcds aren't affected much by the TAA "flickering" at <60 fps

apparently neither is my monitor. I haven't seen any flickering or tearing whatsoever yet using 2t3 or 4t3 AA.

 

[archevilangel]

I don't think it's a gimmick, infact I wish they would transfer over the taa kicking off at a certain fps to regular aa. Think about it, aa when things are running smooth, and no aa when not to make the gameplay more smooth.