ATi is ch**t**g in Filtering

Do u get what the second quote means .

It sounds like to see any diffrence u need a program to take the image of the radeon 9800xt and subtract it by the image of the x800s and then amplifying the result (zooming in alot ?)

which if i understand correctly is a hell of alot more complicated than just blowing up the pics of the image .

Don't know what to make of this .

 

Is it true that there is no correct way to do trilinear filtering? If this was true wouldnt nvidias optimised filtering also be correct, but subjectively not hidding the transitions well enough? I was under the impression that with trilinear virtualy every pixel would be textured with a blend from 2 mipmaps.

As I understand it, you cannot disable this optimisation even when the slider is to the highest quality position. Is this true? If it is, then it appears that no apples to apples comparisons can be made between ATIs and nVidias cards as ATI is always doing its high quality analysed brilinear filtering, and nVidia is doing either worse quality brilinear or full trilinear filtering.

 

So the "Color Mipmaps" setting replaces the color data *before* the textures are filtered?

 

Of course, they don't detect an application, there's no application that coulours mimaps , when there's couloured mipmaps, the sophistication of the algotirhm shows true trilinear

 

Hm.. what I don´t understand is why there seems to be a difference with R420 screenshots in colored mipmap screenshots in areas of tbhe shot where textures are not colored. This is what Demirug spoke about in the 3DC Forum.

I think there´s an easy way to check it. Grab one shot with normal rendering, one with colored mipmaps and compare the areas where textures are not colored. There should´nt be a difference as ATI says the alorithm is based on the data.... and the data in those areas is unchanged.

Lars - THG

 

There is several presentation of info possible:
- One synthetic that assumes that the journalist/reviewer/tester
has made his job of testing correctly (probably exhaustive) and highlights his findings good or bad without flooding the reader with too much charts and so on.
- One exhaustive that just presents figures with little analysis but covering every case so that if a user needs a particular info he will get it by looking at the review.

Drawback of the latter: that implies that the reader knows what he's looking for and that the choice in numbers are not too confusing.
Drawback of the first: people hardly trust a particular reviewer and prefer to make up their mind themselves.

Anyway every way you look at the problem that assumes that the reviewer is honest with the number he presents, a view which can easily be seen as flawed from the start.

 

So then if you were to make all the mipmaps the same color (say, red), the card would always do bilinear filtering and would be FASTER with color mipmaps then without them, correct?

 

ATI is right

ATI's response makes sense. As long as nobody sees the difference, they are entirely within their rights to optimise any features of their graphics cards without telling anybody about it. Even if it means everybody thinks they are getting the full feature, like full trilinear (especially in the light of how they warn people that their competitors are not doing full trilinear). Even if many prominent websites are using colored mipmaps to show the beauty of their filtering patterns, ATI doesn't have to tell us of their optimisation and the fact that it's not what we're getting. What we don't see and don't know can't hurt us. Heck, when I see the colored mipmap patterns on Dave's reviews (like in his X800 XT/PE review), games on my 9600 look even better.

peace

 

Oh, don't be so literal. They already made the "better/worse" explanation earlier in the release. Any difference "affects image quality" in that there are differences, but the point they're saying is that they don't believe it makes the image quality "worse" at all--that it is hardly noticable. (And by all accounts thus far, it is.)

I agree with Hanners, though--there'd better be ways to engage/disengage optimizations like this. People should be able to compare/contrast on their own and pick their preference. I suppose in a way it could be like the angle-dependant AF itself--not something the user has a choice about--but it's better when testing newfangled algorithms and other optimizations to leave the field open.

 

Of course, the algorithm is super sophisticated

 

Patent pending.. and they've had it on the rv350 series for over a year now...

 

They don't address the fact there is a performance drop when coloured mipmaps is enabled.....

Who's ATI trying to fool?

 

The average patent takes like 2 years to go through, I think.

 

i dont buy atis pr. can a driver even do that? reliably detect the similarity between 2 textures that fast?

 

is a patent pending available to the public ?

 

After downloading the 2 .png screen shots I would say that in this particular area the X800 shows better image quality. The texture looks sharper on the X800 whereas it is slightly more blurred on the 9800. I thought maybe I had got them mixed up but I checked from the site again.

If this is the result of the "optimistations" then it appears to be giving better image quality in some (this) situations.

 

If its pending, Surely they have some kind of name for it?

 

Quod erat expectandum.

No one really thought, ATi would go in sack and ashes and confess their sins (if any, that is), right?

But one thing about the explanation is a little bit of a bitter taste to me:
"No-one has claimed that the differences make one implementation "better" than another. "

Does really someone have to go out, whining and complaining, that their texture-filter sucks? I mean, if i say, i see artifacts an incorrectly weighted colors on some pics this does by no means mean, that someone else feels the same about it.

 

Yes they do. Read it again.

That part specifically. No optimizations, means slower.

 

By my standards, it's not cheating, it's an optimization. It's too early to tell whether it's a good one or not, but if you need to use image subtraction combined with amplification to determine the differences then I would say the image quality is the same. The two images may not be identical, but even without optimizations you can get different images from different hardware.

However, on the same token, Nvidia's "brilinear" isn't cheating either: not unless it's described as trilinear.

I think it would be reasonable to allow the user to disable it with a checkbox on the CP (ideally right beside the force trilinear on all texture stages checkbox), but for most people it seems like a decent option.