ATI Filtering Questions

[Quitch]

Just one question, since I seem to have missed something.

Someone (Grestorn) claimed that it had been proved that the ATI filtering method's "adaptiveness" did nothing but detect coloured mip maps. He linked to this article: http://www.computerbase.de/artikel/hardware/grafikkarten/radeon_x800_texturen

I hadn't seen anything regarding this "proof" before now, but I'm seen the "coloured mip maps only" line run more and more recently, which makes me think I missed something.

Can anyone verify what he's saying? What did I miss?

 

[Evildeus]

Hmmm look at the 10+ threads and i think you will have you response. AFAIK, it doesn't detect colored mipmaps it adapts itself in function of the differencies of colors, which gives "true" trilinear in particular cases like colored mipmaps.

 

[Sabastian]

Hmmm look at the 10+ threads and i think you will have you response. AFAIK, it doesn't detect colored mipmaps it adapts itself in function of the differencies of colors, which gives "true" trilinear in particular cases like colored mipmaps.

Yeah and I am getting sick and tired of the whole subject. IMO it is nit picking regarding "trylinier" filtering. While there are legit moralistic objections regarding the whole matter .. this quote sums up my opinion on it.

If you've read most of this feature, you no doubt have noticed a recurring theme: image quality on the GeForce 6800 and Radeon X800 cards is essentially equal. We can empirically show that the anisotropic filtering and anti-aliasing algorithms are different. We can scrutinize extremely blown-up still images and see that the two cards do not, in fact, produce identical pixels. What we cannot do is see instances where one is visibly superior to the other.

Older cards by ATI and nVidia had bigger differences in image quality, but the new series of cards use very similar anti-aliasing and anisotropic filtering sampling patterns. As much as we like that saying, "the devil's in the details," we looked at the details here, and there's just no devil. If you put two computers side-by-side, each running the same game with the same settings, you wouldn't be able to tell which one had a GeForce 6800 and which had a Radeon X800 without looking in the case.

http://www.extremetech.com/article2/0,1558,1607386,00.asp

Move along...

 

[Grestorn]

Hmmm look at the 10+ threads and i think you will have you response. AFAIK, it doesn't detect colored mipmaps it adapts itself in function of the differencies of colors, which gives "true" trilinear in particular cases like colored mipmaps.

It doesn't "adapt". It just turns off the optimizations as soon as the mipmap levels have more than a certain difference. Usually mipmap levels are quite similar, they just use different resolution and therefor level of detail.

The only effect this algorithm has, is to detect mipmap levels that were artificially created to make the filtering visible. Also known as colored mipmaps.

Again, there is no magic "adapting". It's just a way to hide the brilinear filtering, that's all.

 

[Stryyder]

Hmmm look at the 10+ threads and i think you will have you response. AFAIK, it doesn't detect colored mipmaps it adapts itself in function of the differencies of colors, which gives "true" trilinear in particular cases like colored mipmaps.

It doesn't "adapt". It just turns off the optimizations as soon as the mipmap levels have more than a certain difference. Usually mipmap levels are quite similar, they just use different resolution and therefor level of detail.

The only effect this algorithm has, is to detect mipmap levels that were artificially created to make the filtering visible. Also known as colored mipmaps.

Again, there is no magic "adapting". It's just a way to hide the brilinear filtering, that's all.

Whats the basis of your conclusion?

 

[cthellis42]

Then how is "adaptive" AF "adapting," when all it does is have static angle differences?

The point is it changes depending on game, scene, and situation, trying to make the most difference it can without impacting what we can see. It is, in fact, much more "adaptive" than the phrase we already use elsewhere. (And also relating to mip-maps no less.)

This is not to say it will product perfect results, of course--it's still going to make visual differences and be a work in progress. But it's most certainly adapting.

 

[christoph]

@Sabastian
regarding that quote, besides technical interest, i dont give much on how close nv40 and x800 look filtering quality wise when they both are steps backwars in that regard. i dont like it when a years old architecture holds the
crown in this area, i would like to see real progress with new generations. its not that there would be no need for it .

so my hope is the more ppl dig and tell they dont like whats goin on with both ihv's the sooner there may be a change to the better.

 

[Grestorn]

Whats the basis of your conclusion?

It's not *my* conclusion but the conclusion of the author of the aformentioned article. He has gone to great length to determine what the algorithm is doing. He created his own mipmap levels, and continued to change them until the algorithm deemed them to be different enough to disable the brilinear filtering.

The author (Carsten Spille) even described in great detail which criteria a mipmap level must fullfill to be considered "different". And he proved that the algorithm is not adapive at all but just an on/off switch.

Unfortunately, this article is not available in english as far as I know.

 

[Evildeus]

http://www.extremetech.com/article2/0,1558,1607386,00.asp

This article was particularly poor IMO.

 

[Grestorn]

Then how is "adaptive" AF "adapting," when all it does is have static angle differences?

The point is it changes depending on game, scene, and situation, trying to make the most difference it can without impacting what we can see. It is, in fact, much more "adaptive" than the phrase we already use elsewhere. (And also relating to mip-maps no less.)

This is not to say it will product perfect results, of course--it's still going to make visual differences and be a work in progress. But it's most certainly adapting.

We're not talking about the different filtering of differing texture angles. We're talking about the usage of reduced trilinear filtering ranges. Also known as "brilinear".

And the algorithm that enables these "brilinear" optimization only "adapts", in essence, to the fact that the rendering engine uses colored mipmaps or not.

 

[Stryyder]

Whats the basis of your conclusion?

It's not *my* conclusion but the conclusion of the author of the aformentioned article. He has gone to great length to determine what the algorithm is doing. He created his own mipmap levels, and continued to change them until the algorithm deemed them to be different enough to disable the brilinear filtering.

The author (Carsten Spille) even described in great detail which criteria a mipmap level must fullfill to be considered "different". And he proved that the algorithm is not adapive at all but just an on/off switch.

Unfortunately, this article is not available in english as far as I know.

sorry your statement had none of the above qualifications in it I interperted it incorrectly as you knew for certain what the algorithm was doing.

 

[Bjorn]

Yeah and I am getting sick and tired of the whole subject. IMO it is nit picking regarding "trylinier" filtering. While there are legit moralistic objections regarding the whole matter .. this quote sums up my opinion on it.

Move along...

What that article also is saying is that there aren't any noticeable differences between brilinear and trilinear. So i guess all the fuzz about brilinear should be disregarded also, or ?

Look, the problem is that it seems that there are some people that notice more texture shimmering going from the 9800 to the X800. The X800 would have no problems what so ever running full trilinear in any game that's available today. And it's likely that newer games will have other limitations. So why not add an option for it ?

 

[Stryyder]

Then how is "adaptive" AF "adapting," when all it does is have static angle differences?

The point is it changes depending on game, scene, and situation, trying to make the most difference it can without impacting what we can see. It is, in fact, much more "adaptive" than the phrase we already use elsewhere. (And also relating to mip-maps no less.)

This is not to say it will product perfect results, of course--it's still going to make visual differences and be a work in progress. But it's most certainly adapting.

We're not talking about the different filtering of differing texture angles. We're talking about the usage of reduced trilinear filtering ranges. Also known as "brilinear".

And the algorithm that enables these "brilinear" optimization only "adapts", in essence, to the fact that the rendering engine uses colored mipmaps or not.

In the opinion of the article that noone but a few can read correct?

 

[cthellis42]

What that article also is saying is that there aren't any noticeable differences between brilinear and trilinear. So i guess all the fuzz about brilinear should be disregarded also, or ?

It was, if you noticed, by most of the loudest voices complaining now.

Look, the problem is that it seems that there are some people that notice more texture shimmering going from the 9800 to the X800. The X800 would have no problems what so ever running full trilinear in any game that's available today. And it's likely that newer games will have other limitations. So why not add an option for it ?

That would be great. It would also be great if nVidia doesn't--in fact--take away the option to disable brilinear as it's looking like they might in their next driver, but overall We The People are likely to just get screwed.

It would be nice if people complained about all the targets all the time and targetted the right complain points, as opposed to filtering things though their card preferences first. It's a big industry, and we're close to the bottom from a "concern" standpoint. (At least until reviewers get smarter about everything and start reflecting the enthusiast community more.)

 

[jvd]

ATi deserves credit for the fact that the image quality of the cards is not visibly compromised by this filtering; at least no example has yet been seen of this. So far, the brilinear areas have only showed up in laborious tests

http://www.tomshardware.com/graphic/20040603/ati_optimized-13.html

seems toms (who said to be an nvidiot) doesn't see a diffrence unless its done in laborious tests .


With brilinear from nvidia (Which has gotten better but at the start) no one had to find worse case examples because the reduced quality was visable at all times (though that is no longer the case )

with this there is a witch hunt going on and its starting to get very very desprate .

 

[digitalwanderer]

there is a witch hunt going on and its starting to get very very desprate .

I have that weird feeling again that it's gonna get a LOT worse before it blows over too... :?

 

[jvd]

there is a witch hunt going on and its starting to get very very desprate .

I have that weird feeling again that it's gonna get a LOT worse before it blows over too... :?

so do i .

the funny thing is i'm enjoying my sisters x800pro and playing with it the only diffrence i can see compared to my 9700pro si that is much much faster .

 

[Bjorn]

What that article also is saying is that there aren't any noticeable differences between brilinear and trilinear. So i guess all the fuzz about brilinear should be disregarded also, or ?

It was, if you noticed, by most of the loudest voices complaining now.

Was it, who are you talking about then ?

 

[Bjorn]

there is a witch hunt going on and its starting to get very very desprate .

I have that weird feeling again that it's gonna get a LOT worse before it blows over too... :?

I think it's far from a witch hunt and can only lead to things getting better when it comes to honesty from IHV's. And if it get's worse, well, then there's probably a reason for it.

 

[karlotta]

there is a witch hunt going on and its starting to get very very desprate .

I have that weird feeling again that it's gonna get a LOT worse before it blows over too... :?

so do i .

the funny thing is i'm enjoying my sisters x800pro and playing with it the only diffrence i can see compared to my 9700pro si that is much much faster .

what a crock the image on my x800pro sux all the shimmering and banding.. its the worseIQ ever, just AS bad as my 9700pro,AIW9600pro, and my 9800xt. ATI has just stayed the same DAMM them! Oh ya if i look real hard i can get a line on the map... just like all the other cards...... BUT I DOO LUVE 1600/1200 6aa!! whoo whoo.