Good explanation of filtering (must read for beginners)
- Thread starter trinibwoy
- Start date
bloodbob said:
Well, I've followed this issue pretty closely from day one. And I've pretty well read all the posts here at beyond3d on the topic of ATI's adaptive-Tri. I haven't seen any legitimate screenshots that show ATI's adaptive-Tri degrades the mipmaps boundaries on black/white mipmap transitions over what the old Trilinear was providing on the 9800. If you have some links/shots I may have missed I'd like to see them.
bloodbob said:
Look at the shots labeled … xbit
X800 ANIS 8 APP, 45d.
X800 ANIS 16 APP, 45d.
and compare them to the 9800 or the 6800. Those are the shots relevant to the matter at hand. What we need to see are black/white mipmaps (where ATI adaptive Tri takes place) with the X800 set to application-preference-filtering (ie. Full Trilinear on all texture stages). The mipmap transitions on the X800 are as good or better (smoother) than the Tri on the 9800, and better than the Full-Tri on the 6800 in those shots.
bloodbob said:
The whole point of ATI disabling the adaptive-Tri on colored mipmaps is because they knew the adaptive method would degrade IQ slightly. All this shows is how committed ATI is to providing top notch IQ on the X800. They wouldn't even settle for a slight decrease in IQ.
We already had a case of some site forcing ATI adaptive-Tri on color mipmaps to try and show some differences. This is completely brain-dead or downright dishonest. ATI's adaptive-Tri turns off on colored mipmaps as already noted. What's the point of forcing the adaptive-Tri on a colored mipmaps when it doesn't do this.bloodbob said:
The Xbits shots showed very conclusively that ATI's adaptive-Tri does an excellent job on mipmaps -- as good or better than the 9800. So those THG shots look very suspicious. They go completely against what is in the Xbit shots. I think THG's posted bogus shots.
.
bjorn said:
No I don't. The shots are clearly labeled. I've looked carefully at those shots 10x. Every time I reached the same conclusion …X800 Tri = 9800 Tri.
If you have a specific shot where you think the 9800 is doing a smoother mipmap transition than the X800 -- please point it out. And point out … exactly … where in that shot you think the 9800 transitions are smoother.
Blastman said:
I would just like to wonder what proof you have that the screen shots are made up Blastman ( espically since you cross posted this on other forums it must be good proof )?
Anyway if you look at these two images and you look in the inverted section. You will notice how the image goes from sharp to blurry far quicker in one then the other ( look the band near the sent of image ).
http://www.lexicon.net/mccann/1.png
http://www.lexicon.net/mccann/2.png
If you can't work out which is which well thats your problem.
Blastman said:
Take the 4X AF, 0d shots f.e. On the X800, go to the 4th mip map from the "top". You'll see it better on the 6800 brilinear but it's clearly there on the X800 also. This mip map is much smoother on the 9800 and 6800 full tri. Same goes for the "3rd" mip map level.
I did a "blind" test with my very uninterested girl friend and she said 4th from the left, then 2'nd as to which one had the most visible "mip maps levels". As in bri, then try. Not very scientific but at least i'm not alone in thinking that
jimmyjames123
Regular
Some people just have their blinders on, no matter how much you try to reason with them.
COD at a similar position to wherever they used at THG. Using no AF take one shot of bilinear and take one shot of trilinear. Post up the two images let us do a diff. As far as I know on the 9600 and X800 nothing has changed in reference to bilinear/trilinear since either card come out. I have a feeling that he increased the contrast or something so I think I'll also sqaure the difference.
Bjorn said:
Your right Bjorn, the AF-0d shots are "very slightly" worse on the X800. It's difficult to tell because the mipmaps are pushed back on the X800 and that may provide more contrast (ie … smaller area/more condensed). I was looking mostly at the latter shots. On the other hand, in the 45d 8/16-AF shots I see the X800 as better than the 6800. So all in all it doesn't really change my conclusion from that article. The X800 has excellent Tri. Overall (based on that article alone) I would put the 9800 a slight first and the X800/6800 tied a very close second. In any case, I still think that article shows NV's Brilinear is clearly worse than ATI's adaptive-Tri.
bloodbob said:
I have no "proof", just suspicions. THG posted one shot that shows a big difference between the X800/9800's Tri/Bi. All Tom said was ….
"…… This behavior was finally discovered in laborious tests by the Computerbase website. …"
Adding …. "finally discovered" … and …"laborious tests" … just doesn't add credibility to the shot with me. This shot was used to justify his benching the X800 against the 6800 (with Tri-opts-on). Which I obviously disagree with based on what I see in the Xbit article. The xbit article just doesn't show big differences (overall) between the 9800/X800. We obviously need to know the details -- card, drivers, settings, tools …etc…. etc. … of this shot so it can be checked to see what it really means. From what I understand one site has already forced the adaptive-Tri on a color mipmap to see a difference.
Blastman said:
The mipmaps seems to be a bit pushed back compared to the 6800 but not compared to the 9800. And yes, the difference isn't huge. But compare the X800 and 6800 bri in those shots and you could say that the difference isn't huge there either. Then compare bri to full and it's rather big.
I see the same problem in 45d AF 8 and AF16. You can see the "4'th" mip map as a band, not very easy to notice but it's there. Same goes for 30d. There's also not as smoth transition between the 3'rd and 4'th mipmap.
Easiest way to see this is imo to go from bri to try. You can usually spot the same mip map "bands" although the transition between the "bands" is much smoother on the X800. Then compare the X800 to the 9800 and you see a difference immedietly. And ATI is claiming that the quality should be better on the X800 and that seems to be complete BS if these screenshots are valid.
Hmm, I went back, and I did see the transition on the right wall for 4x AF on the 0d shot, where I'd missed it before.
This clearly shows that a difference can be perceived in a still image clearly in a certain circumstance, and it does indicate that 1) "trylinear" was indeed applied for the tests (I wasn't sure before), 2) it's behavior when transitioning to an example mip map level is less gradual than possible, and can't be objectively classified as a "perfect" universal replacement for trilinear when including such a case.
The context of this failure of imperceptibility:
Validating factors (i.e., validating significance of image quality impact towards the blending being significantly visible in general usage)
It is a still screenshot, not an in motion evaluation, and in motion is where the perceptibility needs to be successful. A scene in motion would tend to highlight perceptibility of transitions.
Ameliorating factors
It is a texture of the maximum possible contrast case, and the significant difference occurs specifically where the mip map level being transitioned to is significantly distorted by mip map sampling error, (i.e., where the mip map of the lines fails to resemble the original high frequency line texture, and is undersampled enough to change the data to about 5 line for every 10 that should be present).
...
Given a contrast to the 6800's optimization (which doesn't seem to have been changed from the FX as currently implemented, or at least not sufficiently to be clearly different to what I expect from the FX), which seems to be significantly pereceptible without such undersampling, it does seem to indicate that extra analysis was indeed done for the X800 implementation to suite the output to an analysis of what is required to blend to a lower detail mip map, and it seems successful (for still screenshots) when the mip map being transitioned to doesn't have such gross error.
However, the question still remains on the success of the blending for motion anti-aliasing, and whether a gross error in the lower detail mip map is required for significant perceptibility in that case.
This clearly shows that a difference can be perceived in a still image clearly in a certain circumstance, and it does indicate that 1) "trylinear" was indeed applied for the tests (I wasn't sure before), 2) it's behavior when transitioning to an example mip map level is less gradual than possible, and can't be objectively classified as a "perfect" universal replacement for trilinear when including such a case.
The context of this failure of imperceptibility:
Validating factors (i.e., validating significance of image quality impact towards the blending being significantly visible in general usage)
It is a still screenshot, not an in motion evaluation, and in motion is where the perceptibility needs to be successful. A scene in motion would tend to highlight perceptibility of transitions.
Ameliorating factors
It is a texture of the maximum possible contrast case, and the significant difference occurs specifically where the mip map level being transitioned to is significantly distorted by mip map sampling error, (i.e., where the mip map of the lines fails to resemble the original high frequency line texture, and is undersampled enough to change the data to about 5 line for every 10 that should be present).
...
Given a contrast to the 6800's optimization (which doesn't seem to have been changed from the FX as currently implemented, or at least not sufficiently to be clearly different to what I expect from the FX), which seems to be significantly pereceptible without such undersampling, it does seem to indicate that extra analysis was indeed done for the X800 implementation to suite the output to an analysis of what is required to blend to a lower detail mip map, and it seems successful (for still screenshots) when the mip map being transitioned to doesn't have such gross error.
However, the question still remains on the success of the blending for motion anti-aliasing, and whether a gross error in the lower detail mip map is required for significant perceptibility in that case.
jvd said:
This is not a fact, actually, it has been proven wrong. Look here: http://www.beyond3d.com/forum/viewtopic.php?t=13076
And stop defending ATI! They're here to make money, no matter how, as is nVidia. Blind worshipping of either one just doesn't make any sense at all.
Grestorn said:
Right u just posted that video. Go back and send me the save so I can either back up your claim or deny it .
1 person showing proof and its way after i posted my post about no one doing it.
I hope we get the same witch hunts with the 6800s .... well i guess we need the 6800s before we can witch hunt huh ?