ATi is ch**t**g in Filtering

[CarstenS]

@quasar
you did the same thing (ie bit comparison) between bilinear/trilinear with colored mips for each chip individually as well and compared the generated images between the different chips afterwards, right?

Correct. We also used (that was what Demirug forgot to mention) a pixelcounter for the black dots.

R360 ~310k vs. ~475k (b vs. w)
R420 ~431k vs. 355k (b vs. w)
RV360 ~467k vs. 319k (b vs. w)
(remember, black dots are bad)

edit:
please note, that i am only against black dots in this case! I am NOT a .racist

 

[Dave Baumann]

What application are you using for the image difference (i.e. black and white dots)?

 

[CarstenS]

We used our own tool which, unfortunately is not even remotely in a condition we'd dare it to go public.
Alternatively, you could use photoshop, turning up contrast through the roof. Every Pixel not staying black in this procedure has been changed by trilinear filtering.

 

[dan2097]

Quaser is this behaviour observed just when using control panel/rtool enabled af (i.e. forced af) or when using application enabled af as well?

 

[christoph]

@dan2097
application enabled af as well

 

[CarstenS]

Quaser is this behaviour observed just when using control panel/rtool enabled af (i.e. forced af) or when using application enabled af as well?

A just question, which i admittedly almost overlooked. But christoph's right here - in the last seconds of our time with the X800 we also tried an enabled AF via UT2003.ini.

 

[Bambers]

Do you have the texture preference slider on highest?

I noticed that when I had my 8500 (which was able to do a crude version of brilinear) that in opengl games with a tex pref setting of anything other than high quality it would use its brilinear filter but when coloured mipmaps were turned on it would flip to full trilinear.

with tex pref on highest quality it would use full trilinear regardless.

 

[CarstenS]

Do you have the texture preference slider on highest?

Yes, of course.

 

[MfA]

So are there any benchmarks with color-coded mipmaps and ordinary textures which show the performance increases with ordinary textures?

Oh I see, yes there are Sorry. Quite impressive BTW that ATI managed to get such a linear speedup by adding pipelines as evidenced in those benchmarks, even if it was in a fillrate limited scenario.

 

[Evildeus]

If i understand correctly, Ati forces billinear, and detect mip-map application to make a trillinear picture

 

[991060]

Quasar, have you notified ATi about your interesting finding? I'd like to see their response.

 

[Exxtreme]

If i understand correctly, Ati forces billinear, and detect mip-map application to make a trillinear picture

No. Ati forces "brilinear" filtering. It is a mix between bilinear and trilinear filtering. Here a screenshot:
http://www.3dcenter.de/images/r420_benchmarks/af-test_fx_brilinear.jpg

Left -> full trilinear, Right -> "brilinear" filtering.

And yes, the driver detects whether an application is using coloured mipmaps. In this case the driver enables full trilinear filtering.

 

[L233]

It this is true then ATI stooped to an all-time low. I'd call it malicious consumer deception.

It would also means that most if not all comparative X800 and 6800 reviews done so far are bunk.

 

[CarstenS]

If i understand correctly, Ati forces billinear, and detect mip-map application to make a trillinear picture

No, a reduced-trilinear optimization - by some dubbed "brilinear".

Quasar, have you notified ATi about your interesting finding? I'd like to see their response.

Not yet, but they will receive a pdf-document containing the article.
I'm also very curious about their answer.

edit:
*argh!*
Way to slow again.... This Post may be deleted

 

[BetrayerX]

It this is true then ATI stooped to an all-time low. I'd call it malicious consumer deception.

Careful...green sweat is oozing out of your body.

j/k

 

[991060]

And yes, the driver detects whether an application is using coloured mipmaps. In this case the driver enables full trilinear filtering.

Is it possible to anti-detect this trick without coloring mipmap?

 

[Exxtreme]

And yes, the driver detects whether an application is using coloured mipmaps. In this case the driver enables full trilinear filtering.

Is it possible to anti-detect this trick without coloring mipmap?

I don't know. First it is important to know how the detection works. Then you can think about some counter measures.

 

[Evildeus]

Ok thanks for the clarification *note brilinear*

 

[Hyp-X]

I think the difference between the 9800 and the X800 picture is way too small for trilinear vs brilinear.

What if ATI is doing "fast trilinear"?

If they do that they "have to" detect color mipmaps - otherwise those just show it like it's plain bilinear.

 

[Althornin]

I dont yet see the proof - maybe i'm not understanding something, so spell it out for me, and show me pictures..