Radeon HD5xxx filtering issue

[KARpott]

I tried the search button for a few times but I did not find anything, so I guess I just post straight away.
In the anandtech review of the radeon Hd5870, it says in the AF-section:
"As you can see, the MIP maps in our venerable D3D AF Tester are perfectly circular, the hallmark of an angle-independent implementation. With angle-independent filtering, this effectively marks the end of the filtering arms race. AMD has won, and should NVIDIA catch up in the future the two would merely be tied. There’s nowhere left to go for quality beyond angle-independent filtering at the moment."

I have no idea which AF-Tester you actually used, but take a look at this:
1xAF 2xAF 4xAF 8xAF 16xAF


Settings: Win7 64Bit - HD5850/Cat 10.6 - AF via application - LODBias 0, ALU Rendering perfect - A.I. off - MipMap HQ

I used this AF-Tester: http://www.3dcenter.org/download/3dcenter-filter-tester (sorry, but it's from a German community)

It's also observable in games:
ati: http://www.megaupload.com/?d=6Y5ARKBI
nv: http://www.1filesharing.com/download/0WMNHK0G/TmForever%20gr__n%20B4%20AFx16HQ-2.avi
(please compare the nv-clip with the ati anisotropic clip)
Settings:
ATI: Win7 64Bit - HD5850/Cat 10.6 - 1680x1050 - 4xMSAA via application, box filter - 16x AF via application - ingame maxed out, no postprocessing - A.I. off - MipMap HQ
NV: Win7 64Bit - GTX 275/197.45 - 1680x1050 - 4xMSAA via application - 16x AF via application - ingame setting "best graphics" - driver HQ


Or:
ati: http://h-4.abload.de/img/ut3-amdjk0rr.jpg
nv: http://h-3.abload.de/img/ut3-nvidiajt33y.jpg

Someone else brought this up as well:
http://alienbabeltech.com/main/?p=12648&page=2
Or you might wanna take a look at the thread at 3DCenter
http://www.forum-3dcenter.org/vbulletin/showthread.php?t=482069

EDIT: Look through the first 4 pages to find more information on it. Seems to be a flaw in the AF algorithm. It might just be bug???

 

[Flipped Gazelle]

ati: http://h-4.abload.de/img/ut3-amdjk0rr.jpg
nv: http://h-3.abload.de/img/ut3-nvidiajt33y.jpg

I looked at these 2 pics, and saw that both cards have issues. NV looks worse, though.

Glancing briefly over the ABT article, it looks like the 5770 trades blows with the GTX 285 with regard AF/AA quality. Radeon 4850 is failing on comparison.

 

[KARpott]

If you think so, then you don't get the point. Have you actually watched the videos?
It's not about angle independency, look at those outer rings instead!
Sorry, but that's why ATI hasn't changed a bit for quite a while now. Everyone looks on those stupid AF flowers, but that's not the point anymore. Yeah, ATI IS angle-independent, but the actual filter is terrible.
And the pics you quoted:
Don't you see that half of the texture in the middle is blurred on ati hardware!?! What you see on NV is more moire because IT ACTUALLY DOES FILTER TRILINEAR instead of ATI. With older games, SGSSAA is one possibility to get rid of moire-effects, but you pretty much cannot fully avoid them with texture-filters.
Btw. here are pics with the same AF-Tester on NV-Hardware (gf10x), yeah not "fully" comparable because of different view angle/resolution, but they show the difference I wanted to show, no rings whatever visible here!

-------1xAF -------------- 2xAF ------------- 4xAF -------------- 8xAF ------------- 16xAF ------


GTX460, FW 259.57 - AF via application - LOD 0, ALU Rendering: Perfect - HQ-AF

 

[Flipped Gazelle]

If you think so, then you don't get the point. Have you actually watched the videos?
It's not about angle independency, look at those outer rings instead!
Sorry, but that's why ATI hasn't changed a bit for quite a while now. Everyone looks on those stupid AF flowers, but that's not the point anymore. Yeah, ATI IS angle-independent, but the actual filter is terrible.
And the pics you quoted:
Don't you see that half of the texture in the middle is blurred on ati hardware!?! What you see on NV is more moire because IT ACTUALLY DOES FILTER TRILINEAR instead of ATI. With older games, SGSSAA is one possibility to get rid of moire-effects.

I can't watch the videos where I'm at.

WRT the pics I looked at, again, both cards are messing up. I don't care what the excuse is. If you feel the NV shot looks better, I wouldn't disagree.

In a thread made here a few days ago, some of us suggested that perhaps it's time for video card reviews to include image quality tests, like they did years ago.

 

[Flipped Gazelle]

Btw. here are pics with the same AF-Tester on NV-Hardware (gf10x), yeah not "fully" comparable because of different view angle/resolution, but they show the difference I wanted to show, no rings whatever visible here!

-------1xAF -------------- 2xAF ------------- 4xAF -------------- 8xAF ------------- 16xAF ------


GTX460, FW 259.57 - AF via application - LOD 0, ALU Rendering: Perfect - HQ-AF

Rings are definitely milder on NV, although they are still there. But boy, are there a lot of rendering errors.

 

[aphorism]

"As you can see, the MIP maps in our venerable D3D AF Tester are perfectly circular, the hallmark of an angle-independent implementation. With angle-independent filtering, this effectively marks the end of the filtering arms race. AMD has won, and should NVIDIA catch up in the future the two would merely be tied. There’s nowhere left to go for quality beyond angle-independent filtering at the moment."
BS. every heard of the nyquist theorum? i guess ATi's AF can take 2x infinity texture samples. contrary to marketing, the general consensus is that nvidia's AF is higher quality.

 

[RussianSensation]

I can't wait for BFG to chime in on this. I consider him an expert on these forums with respect to image quality. I will say that AF does look better on NV in OP's post. However, I still think AMD is slightly better in AA quality (Edge-detect, etc.).

 

[KARpott]

Rings are definitely milder on NV, although they are still there. But boy, are there a lot of rendering errors.

I don't see any errors to be honest. Just texture filters struggeling with high frequency textures!

Here are the comparison shots, from TrackMania, the game I made the videos from.
ati: http://www.abload.de/img/tmu-amdjy2ye.jpg
nv: http://www.abload.de/img/tmu-nvidiajk2db.jpg
Nvidia, no rings, ATI visible rings!

Here a few more pics which make me believe that what we see is bilinear AF (Win7 64Bit - Hd5850 - Cat 10.6 / 1680x1050 - 4xMSAA application controlled - box filter - texture filter application controlled - postprocessing deactivated (ingame option) - A.I off)
Bilinear ------------ Trilinear ---------- 2xAF ------------ 16xAF


Another NV pic with everything maxed out and driver HQ


The visible rings on ATI HD5k series are most likely the same rings observed here:

(pic from alienbabeltrech, some transitions marked)
Please don't look at the flower, it says nothing. Instead, test it with high frequency textures.

 

[Flipped Gazelle]

Here are the comparison shots, from TrackMania, the game I made the videos from.
ati: http://www.abload.de/img/tmu-amdjy2ye.jpg
nv: http://www.abload.de/img/tmu-nvidiajk2db.jpg
Nvidia, no rings, ATI visible rings!

Ah, this is better. NV is clearly superior here.

 

[KARpott]

But please be aware that the screens do not show average results, but extreme cases. It, of course, depends strongly on the textures. Lots of textures in TrackMania do suffer from this effect only very slighty, almost impossible to notice while driving. But since textures become more and more highres, those extreme cases appear more and more frequently as well.

EDIT: Anyway, I still believe this can be some kind of cheating, and in my opinion, it's a shame Anandtech calls this AF the best up to now. Nvidia has been criticized for their filters when g70 was out because of their trickery. I think it's now time to do the same with AMD, even though the problem does not seem to be that obvious in most situations. Afterall, the cards do have enough performance to run older dx9 games smoothly, even with sgssaa. So why can't there be a HQ AF filter, that's all I want. I don't care if they keep this as the default quality, I don't even care if they call it "HighQuality", als long as they give us a button to stop the crap demonstrated in the screenshots.

EDIT2:
Here is a better comparison with the AFTester:
first row: GTX460/FW 259.57 - AF via application - LOD 0, ALU Rendering: Perfect - HQ-AF
second row: Win7 64Bit - HD5850/Cat 10.9 - box filter - AF via application - LOD 0, ALU Rendering: Perfect - driver default (CCC)

-------1xAF -------------- 2xAF ------------- 4xAF -------------- 8xAF ------------- 16xAF ---

 

[BFG10K]

I’m the author of the article @ Alienbabeltech you linked to, so I’m well aware of this problem.

ATi underfilters surfaces so they have more aliasing and also visible texture transitions compared to nVidia. These transitions are quite visible in-game during movement.

Super-sampling reduces the texture aliasing, but it can't touch the transitions.

 

[KARpott]

Thanks for your reply, sir.
Do you think the underfiltering happens on purpose?

btw. I fixed the video link for the nv-version in the first post. If you tell me a good host that works for you American guys, I can upload the ati video again.

 

[BFG10K]

Do you think the underfiltering happens on purpose?

The current theory is that it’s a limitation in their texture cache design. Unfortunately ATi isn’t responding to the issue (I emailed Catalyst Maker directly with my findings but I never got an answer), so it’s unlikely we’ll ever know the true reason.

It'll be interesting to see if the 6xxx series fixes the issue.

 

[KARpott]

Thank you, man. I was interested in that kind of information. Thanks a lot. I was sure there's something special going on with the HD5K-filters.

 

[distinctively]

Are there any screen shots where the pics are from identical scenes? Every picture I look at has different angles. It's really hard to make an accurate assumption without better sources.

 

[KARpott]

here's the thread with the direct comparison shots (it's German, but pretty much understandable)
http://www.forum-3dcenter.org/vbulletin/showpost.php?p=8286953&postcount=1259

What you see is that the problem is slightly visible on HD4K as well, but much stronger on HD5K. Several players reported no issues in TM or other games on HD4K, whereas HD5K has several problems in games as well. Nvidia TMUs render almost identical to ALUs.

please forgive the slightly different resolutions
And I think the alienbabeltech article is source enough, what better pics can you have? It's not pixel-counting, it's more like accepting the obvious difference. You can even see that in lower res vids or pics! Hell, you can even observe the problem on youtube vids. Just look for bad content hl2 movies, and you'll notice the transitions.

 

[Anarchist420]

nvidia's filtering much > ATi's

With nvidia there is rarely texture aliasing, with ATi there is a shitload of texture aliasing, which means the experience is ruined with ATi for me.

I don't think ATi has used proper trilinear and decent mip transitions consistently since the Rage 128.

Nothing compares to 3dfx's filtering though.

 

[distinctively]

Okay, I'm really not interested in the shots done from AFtester. It has its value but its still a synthetic test. Also, the different resolutions are creating a different screen ratio. Between resolution and ratio, I'd have to say that these shots can't be considered useful. I just went through Serious Sam HD (which is full of grid-like surfaces) and Wolfenstein (lots of brick roads). I just can't find any trace of what you're showing here. I can't say I've ever noticed a problem on my 5770 which is what I've been using for observation.

Simply, so far I have been enjoying fantastic AF with my current card. My older 9600GT was a great card but its just not in the same league for clarity and overall IQ as my 5770.

 

[MrK6]

Okay, I'm really not interested in the shots done from AFtester. It has its value but its still a synthetic test. Also, the different resolutions are creating a different screen ratio. Between resolution and ratio, I'd have to say that these shots can't be considered useful. I just went through Serious Sam HD (which is full of grid-like surfaces) and Wolfenstein (lots of brick roads). I just can't find any trace of what you're showing here. I can't say I've ever noticed a problem on my 5770 which is what I've been using for observation.

Simply, so far I have been enjoying fantastic AF with my current card. My older 9600GT was a great card but its just not in the same league for clarity and overall IQ as my 5770.

Largely my experience as well. I do notice the filtering edges in some games (there's literally a line across an entire scene where you can see the scene being redrawn under the new filter), but overall the image quality is great. I do hope AMD tweaks this with the 6xxx series, but as long as they keep their amazing 2D image quality, I'll be happy.

 

[Udgnim]

not the same gen of cards, but somewhat related to your post

when I upgraded to a 4870 from a 7950GT, I noticed a difference in filtering (similar rendering issue to your Trackmania comparison shots) in WoW

 

[BFG10K]

Nothing compares to 3dfx's filtering though.

3dfx never implemented any form of AF (Rampage was going to have it), so their filtering is not really comparable to what is happening here.

 

[KARpott]

Okay, I'm really not interested in the shots done from AFtester. It has its value but its still a synthetic test. Also, the different resolutions are creating a different screen ratio. Between resolution and ratio, I'd have to say that these shots can't be considered useful. I just went through Serious Sam HD (which is full of grid-like surfaces) and Wolfenstein (lots of brick roads). I just can't find any trace of what you're showing here. I can't say I've ever noticed a problem on my 5770 which is what I've been using for observation.

Simply, so far I have been enjoying fantastic AF with my current card. My older 9600GT was a great card but its just not in the same league for clarity and overall IQ as my 5770.

Okay, but it's impossible for me to give you equal TrackMania shots since I do not have Nvidia hardware, the ones given in the first post are clear enough. You can't deny that you can clearly see transitions on ATI hardware whereas Nvidia is almost completely transition free. You won't benefit from exactly same scenes in this case!

Once again, my intention was not to call the ATI filters crap, but to point out that they are clearly inferior to Nvidia's filters on current hardware and no review bothers mentioning it. For me, the AnandTech conclusion stated above is simply not true, whether the problem occurs often or not! In most games, I don't notice the lines either, so it might be a rather "synthetic" problem. But it's still there and worth mentioning.

And then: TM clearly does not request a bilinear filter when using AF algorithms. The problem is not caused by the game. It's just the kind of content.
Just high frequency textures alone do not necessarily cause the effect, it's combination of shaders and textures which is the problem. You can see that in the video comparison. The strength of the effect clearly depends on how the scene is shaded.

@UDgnim: As the AFTester results show, the HD4870 suffers from this problem as well, but it's only barely visible! So you might observe this kind of effect with your hardware as well!

 

[Anarchist420]

3dfx never implemented any form of AF (Rampage was going to have it), so their filtering is not really comparable to what is happening here.

I was talking about their trilinear filtering/mip transitions.

 

[SirPauly]

Here is another article about this:

http://www.gpu-tech.org/content.php...tropic-filtering-of-the-Radeon-HD-5000-series

 

[KARpott]

This article ,among other sources like the alienbabeltech-research, quotes the research done by my fellows and me at 3Dcenter forums, although using self-made media. It's pretty new actually!
I hope you're finally convinced that we're really facing a problem here! The more of you stop ignoring (yeah yeah, we all know it only occurs in extreme situations, so you can leave that out now) this, the more likely we will see this fixed in hd6xxx series. TrackMania is one of my favorite games, and I know you don't care because in your game, you don't see banding. But imagine it would be your favorite game being ruined like it is shown in the TM shots.