Radeon HD5xxx filtering issue

[BFG10K]

So play with AF at 2x if the transition bothers you? I mean, it's a racing game, all the textures should be blurred anyway, so what does the level AF even matter?

This is absolutely the wrong attitude to take. If he had an nVidia card he could run at 16xAF because it doesn’t have same problem. Therefore it’s quite reasonable to expect the same level of quality from his ATi board, especially given ATi is in direct competition with nVidia.

If a vendor’s feature is broken in such a fundamental way, you call attention to it and try to get it fixed. Some might find dropping to 2xAF acceptable, but to someone like me that’s been running 16xAF since 2002 in every 3D game I've played, it’s unacceptable.

 

[CitanUzuki]

Looking forward to reading your write up on the 6000 series BFG. Hopefully they get this sorted out.

 

[lavaheadache]

as much as i love my 5870's and the *potential power* they have, I'm rather disgusted with the lack of quality and polish in the drivers. These are AMD's flagship cards and yet they seem to be tended to like a neglected child. Makes me feel like DSS pestering AMD to care for it's product and customers

 

[brandonb]

I just wrote a DX11 app that tests out AF on my ATI 5850, and I'm not really seeing any issue. I also changed it over to the Microsoft Reference driver and it looks almost identical.

Can you tell which side is which?

http://img291.imageshack.us/img291/8239/af16.png

 

[KARpott]

You don't see a difference because the the repetition frequency of the texture is not high enough. Use the same tile texture as in the screenshots above. As I said, this problem occurs in EXTREME cases.
And still, it might be possible that you will not witness the problem, as bilinear filtering + af was disabled by Microsoft in DX11, as far as I know. I have not witnessed it in dx11 games, either. But considering the amount of dx9 games still being developed and being played already, it's still a major issue imo.

EDIT: Here are HL2 screens. Please press once on the pictures (after you pressed on the thumbnails!) to enlarge them; otherwise, the downscaling ruins the result.



Settings: Win764Bit - HD5850/Cat10.9 - 4xMSAA via application, box filter - AF via application - ingame maxed out, bloom activated - A.I off - MipMap HQ

You can play the same scene with these savegames:
http://www.multiupload.com/WB0KBYGV81
I guess there are a lot of NV users with HL2 out there.

Not as obvious as in TM, but clearly visible. See how it depends on the shading. There are textures with way higher repetition frequencies, but this one is lit the right way.

EDIT: And finally, NV-ATI comparison shots (thanks to Gorkon from 3Dcenter):

-left: NV, right: ATI, first two shots: NV: Q, second two: NV HQ-----------------
|

Settings:
ATI: HD5850/Cat 10.9 - 1680x1050 - 8xMSAA via application, box filter - 16xAF via application - MipMap HQ - A.I off - ingame maximum, no postprocessing
NV: GTX 460/FW 259.57 - 1680x1050 - 8xMSAA via application - 16xAF via application - Left: Q,Right: HQ - ingame maximum, no postprocessing

Please note that the advertisement changes with every run of the replay. The rest of the scene is the same.

 

[brandonb]

Hi KARpott,

I will play with it more and see if there is a setting which causes a problem, I can also create the same program in DX9. Then I will change it over to the reference driver and see if the problem happens there as well.

 

[Xarick]

Those shots are interesting. If it werent for that band the filtering would be identical. I wonder if ATI can remove or reduce the band.

 

[BFG10K]

Can you tell which side is which?

http://img291.imageshack.us/img291/8239/af16.png

I think the right side is the Radeon. Still, that scene is just untextured lines, so it's really not that great for testing filtering.

Not to mention that neither side can filter the image without any aliasing, which proves you need super-sampling to do it properly. Even 512xAF implemented on the ALUs can't do it right.

 

[KARpott]

Off-topic:
In case you're wondering. Someone with a bit knowledge in German language was not happy with my previous nick (too scatological). Although my name didn't have anything to do with s..., but instead referred to the German dictionary "Langenscheidt" as well as to the guard from "Hogan's Heroes" (hilarious German sync),I can fully understand that someone could have easily misinterpreted it.
On-topic to brandonb:
Thanks, I hope you like the new comparison shots. I cannot get it any better. And I will deliver some HL2-comparisons as well.

 

[brandonb]

BFG10k, that scene is actually a 6 sided cube with a 2048x2048 line texture on each surface, but I can change the texture.

 

[KARpott]

EDIT:

Here's an NV-ATI video comparison with HL2
Scene is based on this savegame: http://www.multiupload.com/WB0KBYGV81

ATI: http://www.multiupload.com/V1A8H9ZACH
NV: http://www.multiupload.com/UKWVDT3UZI
NV2: http://www.multiupload.com/V6125BMKL0 (the video has a litte compression, unfortunately, but the effect is visible, or not )
Settings:
ATI: HD5850/Cat10.9 - 1680x1050 - 4xMSAA via application, box filter - AF via application - ingame maxed out, bloom activated - A.I off - MipMap HQ
NV: GTX285/GF258.96 - 1680x1050 - 4xMSAA via application - 16xAF via application - ingame maximum - driver HQ+Clamp
NV2: GTX 460/FW 259.57 - 1680x1050 - 4xMSAA via application - AF via application - ingame maximum - driver HQ+clamp

Thanks to Popeye and Gorkon from 3DCenter for the NV videos. I hope the videos work! Try them with different media players. MediaPlayerClassic is the best, imo. Winamp might have problems with HD videos. Be sure to play them in full-screen mode. Otherwise, the media player's downscaling ruins all results.

I also have a NV-ATI comparison shot. Not perfect because of different huds! But the rest is the same. Scene also based on that savegame

Settings:
ATI: as above
NV: GTX 460/FW 259.57 - 1680x1050 - 4xMSAA via application - 16xAF via application - ingame maximum - driver HQ+clamp

Remember, those videos and screenshots show EXTREME CASES only. Of course, everything in these shots has been tailored towards the best possible situation that demonstrates the effect.It's up to you whether you can live with it or not, but there exists an issue, and I don't like it. Next time, I want to know it before buying the card. As we could easily demonstrate, AnandTech and other sites could have found out about that during the reviewing process, because the issue can be observed in simple AF-Testers. Of course, I don't know why this happens, but it looks like there ain't just a driver bug, ATI would have released a fix.

 

[Genx87]

For me the biggest evidence was in Microsoft Flightsim X. On my 4850 I got a lot of popping of textures and terrible artifacts on some angles. Which was completely crappy compared to the 8800GTS 640 it replaced. And my GTX 470 doesnt have this issue either.

 

[KARpott]

Btw. the issue also occurs in OpenGL, even old Q3 engine games suffer from it in certain situations. Example: EliteForce 2
http://www.multiupload.com/B22G3XVSRH
Compressed with H264 codec, but the problem is clearly visible.
Settings:
HD5850/Cat10.9,OpenGL@HD5KReleaseDriver - 1280x1024 - 8xMSAA via driver, narrow tent - AF via driver; trilinear,bilinear via application - ingame maxed out - A.I off - MipMap HQ

I used the tool RadiT 0.2 programmed by Mr. Lolman from 3DCenter to run the game with the OpenGl HD5K-release driver. It is the only reasonable driver I know that enables SGSSSAA for q3 titles. I had to use it because I wanted to show the difference between msaa and sgssaa here.

In case you wanna play with it, here's the savegame:
http://www.multiupload.com/LNTDFTBWF8

 

[Xarick]

Well I never noticed it before you guys pointed it out on TMU. Now I can't not see it. Not that I needed another reason, but selling my card for a 460.

 

[Idontcare]

Thanks to Popeye and Gorkon from 3DCenter for the NV videos. I hope the videos work! Try them with different media players. MediaPlayerClassic is the best, imo. Winamp might have problems with HD videos. Be sure to play them in full-screen mode. Otherwise, the media player's downscaling ruins all results.

Any reason to not use youtube? It handles HD vid up to 1080P (maybe higher, haven't tried more than 1920x1080 myself).

 

[KARpott]

I dunno the maximum bitrate of youtube. What is the resolution worth without a decent bitrate, and those vids have a pretty high one.
You know, I want to avoid statements like "ahh, that's crappy quality, those vids are worth nothing".

 

[brandonb]

Here is a new pic

http://img688.imageshack.us/img688/8239/af16.png

The only way I can get that banding to appear like in the screenshots is to turn off linear mip map filtering, and set it to point filtering.

I wonder if the games shown are not setting the mip map filtering (or its setting it to point), and Nvidia hardware is forcing things to be linear filtering in these cases?

My past experience with Nvidia is that it does default some settings, and causes developers to do things somewhat incorrectly, and the problems manifest on ATI hardware (because it is not defaulting in the same manner.)

(my aircard is fail... Not sure if the pic is even working, I'll upload it again if its not)

 

[BFG10K]

I used the tool RadiT 0.2 programmed by Mr. Lolman from 3DCenter to run the game with the OpenGl HD5K-release driver. It is the only reasonable driver I know that enables SGSSSAA for q3 titles.

Hmmm? I ran SSAA in several Quake 3 games (e.g. Jedi Academy, Call of Duty 1) using nothing more than ATi’s stock control panel.

 

[BFG10K]

I wonder if the games shown are not setting the mip map filtering (or its setting it to point), and Nvidia hardware is forcing things to be linear filtering in these cases?

Interesting theory, but it’s happening in far too many games, both old and new. That and the 4xxx series doesn’t have the same banding in the same games.

Also the tester app only shows it with certain combinations of texture size and AF levels (i.e. within the same app it’s not always happening).

Is there any chance your program is using the ALUs for filtering instead of the TMUs?

 

[KARpott]

Here is a new pic

http://img688.imageshack.us/img688/8239/af16.png

The only way I can get that banding to appear like in the screenshots is to turn off linear mip map filtering, and set it to point filtering.

I wonder if the games shown are not setting the mip map filtering (or its setting it to point), and Nvidia hardware is forcing things to be linear filtering in these cases?

My past experience with Nvidia is that it does default some settings, and causes developers to do things somewhat incorrectly, and the problems manifest on ATI hardware (because it is not defaulting in the same manner.)

(my aircard is fail... Not sure if the pic is even working, I'll upload it again if its not)

Might sound stupid, the your texture is still not suitable. Create VERY small tiles with lots of contrast, or grates (alpha tested or not, I think it does not matter)

@BFG:
Well, sgssaa for eliteforce 1,2 and for rtcw does not work with Cat 10.9 here, but it works with the OpenGL HD5Krelease driver. All guys from 3DCenter have the same problem, the tool has been extended specifically for re-enabling sgssaa in older OpenGL games (that's why the release driver files are included), so I have no idea why it works in jedi academy etc. . My experience is that in Jedi-Academy, alpha tested textures ARE handled, but no REAL SGSSAA is applied.
I think I have to try it out with both drivers to be sure, but yes, my wording was not perfect, I should not have said q3 titles in general. Btw. that tool is fine, an aa and driver profile manager. For each application, you can switch to another driver build and another aa-profile, while the default driver is still the one you have installed.

 

[brandonb]

KARpott, I've tried just about every setting I can think of, repeated the texture 1000 times tried all kinds of different texture types, and I still can't recreate the banding unless I change the mip filtering to point.

BFG10k, I'm not sure, I'm using DirectX11 + the standard HLSL. The settings are all done with the DirectX11 runtime, and I'm using the standard texturing commands in the HLSL that is found in the DirectX11 SDK. I'm not sure what the DirectX11 runtime is doing in these cases.

 

[Tempered81]

I just wrote a DX11 app that tests out AF on my ATI 5850, and I'm not really seeing any issue. I also changed it over to the Microsoft Reference driver and it looks almost identical.

Can you tell which side is which?

http://img291.imageshack.us/img291/8239/af16.png

Reference driver on the right?

 

[brandonb]

ATI is on the right in both the screenshots.

 

[TLDR]

Hello, everyone!

As Radeon 5850 owner, I can confirm the existence of aforementioned filtering problem. I think that hardware limitation is behind it, because "cheating" through drivers in ages old OpenGL games makes no sense.

Brandonb, could you introduce vertical texture movement in your program and compare the results?

 

[Idontcare]

Welcome to the forums TLDR! :thumbsup: