ATi's "brilinear" anisotropic filtering on the RX800

[Exxtreme]

http://www.computerbase.de/artikel/h..._texturfilter/

It is a nice german article. It seems that the RX800 is using "brilinear" anisotropic filtering as a standard setting. Until the driver detects whether an application is using colored mipmaps. In this case the driver disables the "brilinear" filtering. RX800-owners can approve it. With colored mipmaps the fps numbers should be lower.

It means too that almost all reviews are nearly useless.

[bratboy]

Well, I tried to read it anyway but wasnt easy, and I took german in school. I really didnt get what he was going on about that all current benches of x800 and xt-pe were invalid. For some reason I thought "Briliniar" was something that had been around for some time so not sure what the suprise is but maybe Im just too Obtuse to let me change my mind about my purchase of an XT-PE.

Bilinear, Trilinear, Brilinear or "Hyper uber take that Linear" I dont care as long as it performs well. Its my satisfation not some sites or self appointed Guru's opinion that matters in the end. I am more than certain that I will recieve as much if not more satisfaction from my new card than I pay for it so to me its good value

[Mojoe]

Old news.

http://www.beyond3d.com/forum/viewto...=12153&start=0

[Kombatant]

Originally posted by Exxtreme
http://www.computerbase.de/artikel/h..._texturfilter/

It is a nice german article. It seems that the RX800 is using "brilinear" anisotropic filtering as a standard setting. Until the driver detects whether an application is using colored mipmaps. In this case the driver disables the "brilinear" filtering. RX800-owners can approve it. With colored mipmaps the fps numbers should be lower.

It means too that almost all reviews are nearly useless.

The X800? I thought that if you forced AF through the CP, it happens in every Radeon card Only if the app requests full trilinear and the slider is in app pref you get full trilinear.

If that's what the article is about, we are talking about one year+ old news

[Quasar]

Well, to put it straight it's basically all about brilinear filtering similar to what the GeForce FX was bashed to death for last year.

Problem ist, if you enable AF in the control panel, you get Texture-Stages Optimization. To circumvent this, a lot of people use the rTool - which works perfectly in all Catalyst releases up to an including 4.5.
But ONLY on R3xx Chips.
RV3xx and R420 do switch off the Textur-Stage Optimization but still retain their brilinear filter - no matter what you switch on or off or which tool you use.
Also with application driven AF, as in UT2003 possible by altering the ut2003.ini, brilinear remains active.

edit:
Sorry, but online translators just plain suck!

[Exxtreme]

Kombatant,

it is NOT the texture stage optimisation.

Here a screenshot:
http://www.forum-hardtecs4u.com/3dce..._brilinear.jpg

Left side-> trilinear, right side -> "brilinear". It is a mix between bilinear filtering and trilinear filtering. It saves a lot of texel fillrate but it reduces the image quality. And the strange thing is that the driver detects if colored mipmaps are in use.

[Kombatant]

Since my Deutsch is extremely rusty and I doubt i will understand many things from the article, one question: does this mean that when you have AF to App. pref., and the game requests full trilinear, it won't get that?

[Quasar]

Yes, same thing as with nV-Brilinear...

[Exxtreme]

Originally posted by Kombatant
Since my Deutsch is extremely rusty and I doubt i will understand many things from the article, one question: does this mean that when you have AF to App. pref., and the game requests full trilinear, it won't get that?

Exactly.

[spyre]

Yep seems so, you can make UT2003/4 request full stage Trilinear AF and thats one of the games they tested

Coloured mip maps and the results changed back to normal none brilinear

[ellingsen1]

I don't understand German or the translated version very well but if I understand you guys right even if you use the application for setting the filtering they turn on all texture stages but still only use brilinear? They lowered the image quality from the R3xx series? That would be the same thing Nv did last year. Doesn't NV now have a way to turn off the brilinear filtering now to give you full quality? If that's the case it does not say much for the worth of some of the benchmark scores out now.

[spyre]

It could be that the drivers dont work with the force full stage trilinear option and R420 cards but that doesnt explain why they found that the results differed when coloured mipmaps were used instead of normal ones

[vaaran]

one of the most questionable things with this seems to be the driver detects when coloured mipmaps are used ans switches the brilinear optimization off

[ellingsen1]

"Coloured mip maps and the results changed back to normal none brilinear"

If that's the case ATI has sunk to the same level as NV. Nothing like trying to pull the wool over the consumer and reviewers eyes. It would have been one thing just to use it but it's another to deceive with detection of colored mip-maps and make you think you are getting something when your not.

[Blacklash]

Honestly guys it wouldn't phase me if ATI was doing this, I think a much large deal has been made of brilinear than is warranted. With the naked eye I don't see differences. Of course with the color analyzer I do, but I don't play my games with the analyzer up

[Exxtreme]

Originally posted by Blacklash
With the naked eye I don't see differences. Of course with the color analyzer I do, but I don't play my games with the analyzer up

Please try to play Tron 2.0 with "brilinear" filter. I swear, you will see the difference.

[ellingsen1]

Originally posted by Blacklash
Honestly guys it wouldn't phase me if ATI was doing this, I think a much large deal has been made of brilinear than is warranted. With the naked eye I don't see differences. Of course with the color analyzer I do, but I don't play my games with the analyzer up

To me it's not the fact that they are doing brilinear but the fact that the went to great lengths to hide it. It also skews some benchmark results because many reviewers forced the NV cards to use no optimizations to make it more what was thought to be apples to apples. That would make a decent diff. when using higher resolutions with AA/AF.

[spyre]

To be slighty fair even the disable option in the Nvidia drivers doesnt disable it completely

[Blacklash]

Originally posted by ellingsen1
To me it's not the fact that they are doing brilinear but the fact that the went to great lengths to hide it. It also skews some benchmark results because many reviewers forced the NV cards use no optimizations to make it more what was thought to be apples to apples. That would make a decent diff. when using higher resolutions with AA/AF.

I agree with you here. If it's going to be brilinear then the NV cards should be allowed to run that. Many sites included benches with both. Looks like they need to remove the "opt. off" results.

[tEd]

deja vu

[ellingsen1]

There is a post at B3D by jimmiejames here that shows what I am talking about benchmark wise. It shows the NV card clobbering the ATI card with just 4xAA and then the ATI card only losing a couple fps once AF is added enabling them to beat NV. I guesss I now know why the reviewers were calling ATI's AF almost free.

http://www.beyond3d.com/forum/viewtopic.php?t=12486

[Dave Baumann]

Exxtreme / Quaser - what did you to to determine there "Brilinear" patterns enabled? Just look at difference images between 9800 and R420? If I do that, I can certainly see there are differences between the two, but then there are still differences if I do an image compare with coloured mipmaps.

[Quasar]

Originally posted by DaveBaumann
Exxtreme / Quaser - what did you to to determine there "Brilinear" patterns enabled? Just look at difference images between 9800 and R420? If I do that, I can certainly see there are differences between the two, but then there are still differences if I do an image compare with coloured mipmaps.

Just wanted to reply to that question over at B3D, but i can't reach the forums right now.

Anyway, we used a bit-correct comparison, where the color/brightness values of each pixel was compared and, if they were the same, the pixel was colored black.
This was done with the regular images from each card taking a 4xAA/16xAF-BILINEAR Shot and a 4xAA/16xAF-TRILINEAR (forced trilinear, not TS-optimized).

[Dave Baumann]

Originally posted by Quasar
Just wanted to reply to that question over at B3D, but i can't reach the forums right now.

Anyway, we used a bit-correct comparison, where the color/brightness values of each pixel was compared and, if they were the same, the pixel was colored black.
This was done with the regular images from each card taking a 4xAA/16xAF-BILINEAR Shot and a 4xAA/16xAF-TRILINEAR (forced trilinear, not TS-optimized).

Yes - I do a bit comparison of R360 and R420 with just trilinear I see differences. However, I also see differences with a bit comparison when the mipmaps are coloured. Did you do that test?

[Quasar]

I think, we're talking in different directions...

What we did, was comparing the differences between R420 bi- and trilinear AF on one picture and R360 bi- and trilinear AF on the other Picture (RV360 is just for completeness' sake).

Or did you mean it just that way?

[Dave Baumann]

Thats not a complete test though.

Try doing the same thing with mip colour enabled and see if you still see differences between R360 and R420. At the moment you may only be comparing the differences between how the two architectures handle different mip map transitions.

Its been mentioned that when you turn "mip colouring" on you can't "see" that brilinear is enabled, as it just looks like full trilinear ergo you reach the conclusion that the driver is detecting it and turning off the optimisations. However if its the case that the R420 drivers are detecting that mip colouring is on then there should be no difference between a comparison shot of R360 and R420 with mip colouring on, however this is not the case with images I've just done (unless I've done something wrong).

[Quasar]

Originally posted by DaveBaumann
At the moment you may only be comparing the differences between how the two architectures handle different mip map transitions.

I hope we do not misunderstand each other, but exactly that is what it's all about. Do both chips calculate from one mipmap to the next linearly with LOD or does, at a given LOD-Value, suddenly a clean bilinear mipmap interpolation kick in?

If you insist on comparing different chips you'll of course find differences between R360 and R420 even with bilinear filtering applied, heck also RV360 and R420 do slighty different computations.

[Dave Baumann]

Again - run same test with the mip map colouring enabled. You will see the same "bands" between the mipmap transisitions as you saw without the colouring (except the coloured shot difference will be accentuated); however according to whats been said the "optimisations" are turned off when mip-map colouring is on, in which case the bands that are seen with the non-coloured images wouldn't be seen on the coloured images (and yet they are).

[ellingsen1]

I feel a little better about ATI after reading DB's posts. From his posts it sure sounds like he's trying to say that there was no intent to deceive by detecting the use of colored mip-maps. That does not change the fact that it appears that when the application asks for trilinear filtering they are only providing brilinear. My comments on the skewed benches still seems valid.

[Quasar]

Since our sample had to go back to the manufacturer, i can only offer you the shots we've already taken:

One and Two.

As you can see, the differences are as small as in the regular R9800 shot (mainly stemming from moving particles and clouds in the background).

edit:
/me stupid. /me cannot type url-tags