ATi is ch**t**g in Filtering
- Thread starter Bitpower
- Start date
Mark0 said:
either have i . thats why i was asking . I posted a video too www.ooakalabeth.com/video its a multiplayer map.
I can't tell with the videos
DegustatoR
Legend
OK, the screenshots.
First comparision.
Second comparision.
Third comparision.
Forth comparision.
All are 24 bits PNGs, around 150 KB. I plan to include FX5900, R9800 and GF6800 screenshots in the nearest future.
Look for noise on the textures (very noticable on the floor texture in second comparision), though it's definately hard to see on a static screenshot, only when comparing to a full trilinear quality of GF4. And do remember that all the noise and shimmering starting to be VERY bad when you really play the game.
P.S. 2 Hanners: Latest version of rTool doesn't allow to set aniso types (perf/quality/q-app) on r420, only the level of aniso.
First comparision.
Second comparision.
Third comparision.
Forth comparision.
All are 24 bits PNGs, around 150 KB. I plan to include FX5900, R9800 and GF6800 screenshots in the nearest future.
Look for noise on the textures (very noticable on the floor texture in second comparision), though it's definately hard to see on a static screenshot, only when comparing to a full trilinear quality of GF4. And do remember that all the noise and shimmering starting to be VERY bad when you really play the game.
P.S. 2 Hanners: Latest version of rTool doesn't allow to set aniso types (perf/quality/q-app) on r420, only the level of aniso.
I don't think a comparison between the GF4 and X800 really says much about the X800's use of "brilinear" filtering. It's been shown many times before that the R3xx archtecture is more likely to show texture aliasing in cases like those you just posted anyway.
In other words, all you're showing is that the NV2x is better at texture filtering than the R3xx, which we already knew, but not anything about the R420's pecularities.
In other words, all you're showing is that the NV2x is better at texture filtering than the R3xx, which we already knew, but not anything about the R420's pecularities.
DegustatoR
Legend
kyleb
Screenshots are lined up in one pixel precision. You can cut them, put one on top of another in Photoshop or similar program and see for yourself by switching off and on top layer.
Off course I can't line up parts with in-game animation like Max's head for example.
Chalnoth
Since when do R3x0 filtering are considered to be worse than GF4's? Note that i'm talking about trilinear+aniso on angles where R3x0 are doing full aniso.
And, IIRC, i hadn't seen this kind of alaising on any of my Radeons (9600, 9600Pro, 9600XT, 9700Pro, 9800Pro, 9800XT if i'm not mistaken). with an except of R9000Pro, which doesn't do trilinear with aniso at all.
But that's a good point. That's why i'm planing to include screens of the same scenes from R9800Pro, FX5900 and GF6800 Just have to learn to inject coffee directly into my brain...
Screenshots are lined up in one pixel precision. You can cut them, put one on top of another in Photoshop or similar program and see for yourself by switching off and on top layer.
Off course I can't line up parts with in-game animation like Max's head for example.
Chalnoth
Since when do R3x0 filtering are considered to be worse than GF4's? Note that i'm talking about trilinear+aniso on angles where R3x0 are doing full aniso.
And, IIRC, i hadn't seen this kind of alaising on any of my Radeons (9600, 9600Pro, 9600XT, 9700Pro, 9800Pro, 9800XT if i'm not mistaken
). with an except of R9000Pro, which doesn't do trilinear with aniso at all.But that's a good point. That's why i'm planing to include screens of the same scenes from R9800Pro, FX5900 and GF6800
Just have to learn to inject coffee directly into my brain...D, GF4 has better filtering than anything ATi has put out at least since the 8500 b/c:
a) 8-bit vs. 5-bit interpolation
b) "full" vs. angle-dependent AF
The 8500 was known to have more texture shimmer than the GF4. That shimmer was either caused by lower LoD or by the lower-precision interpolation, and the latter probably explains the "noisier" textures in your pics. Or maybe nV uses a better mipmap minimization algorithm?
a) 8-bit vs. 5-bit interpolation
b) "full" vs. angle-dependent AF
The 8500 was known to have more texture shimmer than the GF4. That shimmer was either caused by lower LoD or by the lower-precision interpolation, and the latter probably explains the "noisier" textures in your pics. Or maybe nV uses a better mipmap minimization algorithm?
DegustatoR
Legend
Well, i don't see much more jugginess on my R9800 compared to my Ti4600...
I repeat: i have these cards, so i don't have to go and search for screenshots i can make them myself.
As for low precision color interpolation -- i'm sure that it shouldn't produce such terrible results on R420, there's definately something wrong with current LOD selection algos.
And look here: http://www.computerbase.de/news/hardware/grafikkarten/2004/juni/volle_filterung_ati_nvidia
If this is correct, then i can disable ATI's trilinear optimisations and check if that was the reason of texture noises in Max Payne 2 and Halo...
I repeat: i have these cards, so i don't have to go and search for screenshots
i can make them myself.As for low precision color interpolation -- i'm sure that it shouldn't produce such terrible results on R420, there's definately something wrong with current LOD selection algos.
And look here: http://www.computerbase.de/news/hardware/grafikkarten/2004/juni/volle_filterung_ati_nvidia
If this is correct, then i can disable ATI's trilinear optimisations and check if that was the reason of texture noises in Max Payne 2 and Halo...
DegustatoR said:Well, i don't see much more jugginess on my R9800 compared to my Ti4600...
I repeat: i have these cards, so i don't have to go and search for screenshots
As for low precision color interpolation -- i'm sure that it shouldn't produce such terrible results on R420, there's definately something wrong with current LOD selection algos.
And look here: http://www.computerbase.de/news/hardware/grafikkarten/2004/juni/volle_filterung_ati_nvidia
If this is correct, then i can disable ATI's trilinear optimisations and check if that was the reason of texture noises in Max Payne 2 and Halo...
remember to turn off aniso and to turn on trilinear. ANiso hides the shimmering .
You will see alot on the 9700pro and x800s if u turn of aniso.
I have never played with out aniso since my 8500pro though.
No, jvd. The differences are clearly visible with anisotropic enabled as well (though without anisotropic the R3xx just has horrid shimmering). The differences aren't visible in every texture, but they are there.
Some clear differences I've noticed:
1. NWN: some stone textures have black borders around light grey stone. This is a prime place for texture aliasing, and the black borders around the stones look much more jagged on the R3xx.
2. Serious Sam: SE: has some very similar textures that show the same artifacts as NWN.
3. Everquest: Mostly I notice the problem on specific hills in the game. In particular, the hills that border the Dreadlands seem to show shimmering artifacts.
For me this filtering difference has always been somewhat of a, "I don't notice it often, but when I do notice it, it jumps right out at me."
This is what I don't want: graphical glitches that distract me from the game. Aliasing is one such glitch.
Some clear differences I've noticed:
1. NWN: some stone textures have black borders around light grey stone. This is a prime place for texture aliasing, and the black borders around the stones look much more jagged on the R3xx.
2. Serious Sam: SE: has some very similar textures that show the same artifacts as NWN.
3. Everquest: Mostly I notice the problem on specific hills in the game. In particular, the hills that border the Dreadlands seem to show shimmering artifacts.
For me this filtering difference has always been somewhat of a, "I don't notice it often, but when I do notice it, it jumps right out at me."
This is what I don't want: graphical glitches that distract me from the game. Aliasing is one such glitch.
DegustatoR
Legend
Screenshots above are done with max aniso available (8 for GF4, 16 for R420). It's also safe to assume that MP2 uses trilinear when you select "Anisotropic" in its launcher.
I see little point in comparing only trilinear since nobody plays without aniso these days. But i do have same scene screenshots with pure trilinear, without aniso, just in case Their quality is very close between all cards tested. No shimmering on R420, nor on FX5900. All the noises appears only when you switch to aniso filtering in the game launcher.
I see little point in comparing only trilinear since nobody plays without aniso these days. But i do have same scene screenshots with pure trilinear, without aniso, just in case
Their quality is very close between all cards tested. No shimmering on R420, nor on FX5900. All the noises appears only when you switch to aniso filtering in the game launcher.jvd said:DegustatoR said:Well, i don't see much more jugginess on my R9800 compared to my Ti4600...
I repeat: i have these cards, so i don't have to go and search for screenshots
As for low precision color interpolation -- i'm sure that it shouldn't produce such terrible results on R420, there's definately something wrong with current LOD selection algos.
And look here: http://www.computerbase.de/news/hardware/grafikkarten/2004/juni/volle_filterung_ati_nvidia
If this is correct, then i can disable ATI's trilinear optimisations and check if that was the reason of texture noises in Max Payne 2 and Halo...
remember to turn off aniso and to turn on trilinear. ANiso hides the shimmering .
You will see alot on the 9700pro and x800s if u turn of aniso.
I have never played with out aniso since my 8500pro though.
Why would you turn Ansio off? The whole point of Ansio is to hide shimmering. Telling us there's shimmering without Ansio is hardly a revelation. I play with 16AF in every game, the reason is to hide shimmering. If it happens with 16AF then this is something to worry about, but I couldn't care less about no AF and I doubt anyone else does... texture aliasing is the whole reason for AF exisiting, why turn it off only to prove this point?
But what does that matter if Ansio deals with the problem? It's no use pretending that filtering is an issue alone, it isn't and never will be. If the shimmering is worse but eliminated entirely by AF, or pushed far enough back that you can't notice the difference, then it doesn't matter.
This is what should be being tested. Finding out that the shimmering is worse is a fairly meaningless test without any Ansio testing, unless you're saying that non-Ansio tests of this nature can be extrapolated to an Ansio situation.
I'll stop saying Ansio now
This is what should be being tested. Finding out that the shimmering is worse is a fairly meaningless test without any Ansio testing, unless you're saying that non-Ansio tests of this nature can be extrapolated to an Ansio situation.
I'll stop saying Ansio now
If you're implying that the FarCry video of the x800 are made with AF off, you're dead wrong.jvd said:
Another effect of 1xAF (no AF) is that textures become blurry in the distance (mipmap filtering). A very obvious effect which should prove that neither of the videos were created with AF off.
Similar threads
- Locked