[Tamlin]

I followed this article at alienbabeltech and tried it out for myself as well:
http://alienbabeltech.com/main/?p=12648&page=2

With D3D AF tester, I get grey rings that grows the more I use AF and the chessboard pattern dissappear.

Colored mipmaps seems perfect.

When I used 3Dcenter Filter tester in split mode, the ALU rendering was very good without grey rings, while the TMU had grey rings. Can anyone explain what is going on? Click on resize to see it less compressed.

[CosmoKramer]

I don't have anything to add other than yes - it is broken. The shimmering is noticeable in pretty much every game I play

I recently upgraded from an old 8800 GT which had much better AF (although much worse AA...apparently we can't have it all...)

[Tamlin]

Seems that there is no change going from 8X AF to 16X AF in 3DCenter Filter tester either. Here's a link to the test if someone else wants to try:
http://www.3dcenter.org/3dtools/filter-tester-en

[Space Giraffe]

The grey transitions are definitely various levels of blurring. They seem to be done with bilinear? I don't notice it on most surfaces but on grates and what not the transition can be annoying. I noticed it on my HD 4670 as well. Seems ATI has been using such optimizations for a while.

http://www.hardocp.com/images/articl...vJfn_6_2_l.png

[no-X]

In fact RV7xx GPUs aren't affected.

[Arnold Beckenbauer]

You all are n00bs!
It's not broken, it's optimized.
R520's AF quality was broken, because it delievered really good filtering quality without funny samples savings. DAAMIT has reduced the filtering quality with R600 - RV700, but with Evergreen it's a little bit better (~ Nvidia's Quality Level with all optimizations or so).
The ugly thing is: When Novum (he has written 3DC Fillrate Tester) tried to convince with animated PNGs that RV700's filtering quality is a joke, OpenGL Guy's reaction was really strange ("GIFs had a bad quality").

[no-X]

I think evergreen is better in terms of angles and shimmering (less moire), but it creates strange breaks between some mipmaps. It's visible in games, which use bump-mapping techniques. I hope that R9xx will fix it.

[Space Giraffe]

Originally Posted by no-X

In fact RV7xx GPUs aren't affected.

It isn't as obvious on the R600-RV770 GPUs since they don't use so much blurring but the bilinear looking transitions are there. The image I linked shows it actually.

[no-X]

Blurring? Absence of moire isn't result of blurring, but a result of better filtering (denser samples).

Do you have any game screenshot, which would show bilinear-like transition with RV7xx? These theoretical test doesn't reflect real behaviour in games in many cases. I know RV8xx is able to produce under specific circumstances, but I have never noticed visible break on RV7xx.

[ninelven]

I don't have either one installed anymore, but it was pretty evident in both Guild Wars and WoW. It seemed to be most noticeable in green and orange textures for whatever reason.

[Space Giraffe]

Originally Posted by no-X

Blurring? Absence of moire isn't result of blurring, but a result of better filtering (denser samples).

Well the appearance on textures with the grey zone is blurring. But I agree that it's better filtering regardless of what exactly they're doing.

[hoom]

So is there any kind of official explanation for this?

On the one hand, the big area of grey & hard transitions looks pretty bad.

On the other hand, the moire kicks in further back on the 5870 pic & the grey is hiding a bunch of moire so could arguably be a good thing?

[no-X]

I think lack of moire is good thing (less shimmering), but the hard transition is the problem. I'm not sure if this behaviour is a feature (related to halved texture-caches) or a bug.

I asked Dave some time ago - no response... Local reviewer asked AMD's PR - no response...

Maybe AMD wants to keep it back till the launch of R9xx. There were almost no complains for more than 3 quartals, so it's probably not a serious issue for majority of users.

[Novum]

That's a clear case of "brilinear" filtering. It's a performance cheat.

ATI is doing a heuristic detection in the driver to decide whether or not to use brilinear instead of real trilinear filtering. There is no option to turn this of, but to disable A.I. which will disable all app optimizations and cripples CrossFire to the point of uselessness.

There is also no way to disable "sample optimizations" on Radeons which cause texture flickering, because of undersampling of the line of anisotropy. R600 to RV770 were especially bad, RV870 takes more samples, but still not enough. Since R600 you can't get perfect sampling even if you disable A.I.

So yes, ATIs AF is broken and there is no way to get perfect quality on a 500$ board.

[CarstenS]

Very early R600 drivers still allowed decent filter quality.
There's a nice place in-Game to show this in HL2. I hope i can post it this evening.


edit:
Here it is on Hd5870: http://www.youtube.com/watch?v=uYd_LLRbGHU.
I'll try and add another one with a 4870 from the same spot.

edit #2:
same spot with HD 4870:
http://www.youtube.com/watch?v=Z_vICZsMTQE
This is NOT about the shimmering - that's the fault of the tunnel's texture/shader being high frequency itself and should be hurting the eye on any texture filter out there.

[swaaye]

I haven't seen very many people talking about the filtering quality being poor. I don't remember seeing reviewers complain either. Regardless though, it's not even remotely as bad as what NV had with their pre-G80 hardware (but not many seemed to notice that either).

[caveman-jim]

the angle dependence is better (supposedly now independent, if it's not I can't tell) but there are some optimizations going on that cause shimmer etc. in games etc.

I mentioned it briefly here but haven't heard anything from inquires made. TBH you have to be really looking for it.

[Novum]

Originally Posted by swaaye

I haven't seen very many people talking about the filtering quality being poor. I don't remember seeing reviewers complain either. Regardless though, it's not even remotely as bad as what NV had with their pre-G80 hardware (but not many seemed to notice that either).

All we ask for is a simple option to disable all filtering optimizations in Catalyst. It's not like that's a huge effort to do.

They just don't want to do it, because it makes them look worse in benchmarks. Especially for Crossfire setups it's ridiculous that you have to look at texture banding and flickering despite having >100 FPS in almost all games. I can't stand it to buy expensive hardware and being treated that way as a customer. They can play their games with other idiots.

Speaking of, they also don't publish the registers for setting anisotropic texture filtering in their open specs. What a coincidence.

[madyasiwi]

Originally Posted by CarstenS

Very early R600 drivers still allowed decent filter quality.
There's a nice place in-Game to show this in HL2. I hope i can post it this evening.

edit:
Here it is: http://www.youtube.com/watch?v=uYd_LLRbGHU.
I'll try and add another one with a 4870 from the same spot.

It reminds me of similar mix of dark/night scenery, grainy road texture, and car headlights in NFS Carbon -- which makes the (partial) shimmer on HD 4K series very well pronounced back then. I wonder what would be the case with 5K series in this game.

[CarstenS]

It's not the shimmering in this case - this tunnel in half life 2 is "evil content" which should shimmer/flicker on any reasonably implementable texture filter. It's the banding in the part of the light cone a bit farer away.

Here's the same scene with my HD 4870, btw.
http://www.youtube.com/watch?v=Z_vICZsMTQE
You'll notice the banding has gone.

If youtubes HD-flicks are not enough, i probably can upload the original avi-files somewhere (just need to find a decent host for it).

Originally Posted by swaaye

I haven't seen very many people talking about the filtering quality being poor. I don't remember seeing reviewers complain either. Regardless though, it's not even remotely as bad as what NV had with their pre-G80 hardware (but not many seemed to notice that either).

We complained, and were called fanboys for it. Not by AMD of course.

[fellix]

Could you capture the same scene with some SSAA -- 4 and 8 samples?

[CarstenS]

Yeah, sure (tonight, hopefully after we've entered the world cup finals ) - since i played it with that setting also, i can tell you that it helps At least it's moving the banding further away.

For good manners, here's what a GTX 280 does at driver default
http://www.youtube.com/watch?v=6kTxzCZbV8M

[Jawed]

The absolute silence of AMD in this thread's pretty deafening, huh?

Since NVidia cards are better, video comparisons should be a slam-dunk. I suggest www.mediafire.com for original-video uploads.

Jawed

[CarstenS]

Thanks, I'll take a look at it when back from work.
Meanwhile the last vid, I've captured this morning already.
This time the GTX 280 at HQ (which doesn't against the concentrated evilness of this tunnels' content)

http://www.youtube.com/watch?v=PHbxkyPH1BY

[Novum]

Originally Posted by caveman-jim

the angle dependence is better (supposedly now independent, if it's not I can't tell)

Angle dependence is a non-issue since R5xx. You can only see the difference between the reference rendering and RV770 with colored mips. If you can't tell them apart in this synthetic case there is no way this is visible in games.

The perfect angle independence of RV870 is just a red herring to distract from the real problems.