How long has ati been cheating for?

[the_archer]

Just how long have ati been cheating for? This is appalling. ati has acted like its superior and all this time its been duping everyone. But does anyone know when they started this stuff? Which driver build? I mean how long has stuff been going on for? :lame:

[jAkUp]

They said the filter optimizations started with the 9600xt. But honostly, looking at the screenshots and stuff.. I still think ATI has superior image quality... It just looks sharper.. I know we are looking for trilinear quality here, but the bottom line is iq right?

http://www.reflectonreality.com/nv40r420/

[the_archer]

If they had just quietly been doing this stuff it would be one thing but the fact is that they took a holier than though attitude and touted these cheats as an advantage. Thats wrong just plain, flat out wrong.

[jAkUp]

Well, after looking at those IQ comparisions again, and again... I actually think I changed my mind, certain parts look better on the nV card, such as the blending between mipmaps (trilnear filtering) While other parts look better on the ATI card, such as texture sharpness.. toss up for me.

[Kamel]

personally i think IQ is over-rated. i don't notice the difference between ATi and nvidia cards on nearly anything, even when zoomed in 100 times, i have no freaking clue what they are trying to get at.

i do know that i think nvidia looks better... probably just settings or something, but i remember at a lan party recently even though i had better frame rates, i was staring at the scren of the guy next to me because the game (unreal tournament 2004) looked much better on his screen than mine.

/me shruggs

like i said, i honestly don't see much of a difference in IQ. what little there is i feel is honestly psychological (between ati/nvidia). if you were to go up to 2 machines not knowing what cards were in them and just saw a rendered scene (freelook), could you honestly tell me which one was ATi and which was nvidia by the image quality alone? (framerates capped and everything else, to make sure there's no difference between them other than IQ)

[fivefeet8]

The filtering optimization has been in the 9600 cards for a while now. It wasn't used in the 9700/9800 lines. Both companies optimize. Nvidia uses Brilinear, well atleast there is an option to disable it in the Nv40 now, and ATi used a Control Panel(disabled when using per application settings) optimization that not a lot know about and now it's "trylinear"(cannot be disabled) optimization as well.

I think IQ is a very subjective thing. So having an option to disable optimizations is a good thing.

[Blacklash]

There are rational positions on both sides of the fence. I have read all the threads available.

This article pretty much sums up how I feel about the issue.

http://www.techreport.com/etc/2004q2/filtering/index.x?pg=1

[SmuvMoney]

If they had just quietly been doing this stuff it would be one thing but the fact is that they took a holier than though attitude and touted these cheats as an advantage. Thats wrong just plain, flat out wrong.

Outside of the misrepresentation - intentional or not - in the PR and marketing docs, how is the adaptive filtering algorithm bad or a "cheat"? So far all I have seen is equivalent IQ between this adaptive filtering and the full legacy trilinear filtering (COD shots above nonwithstanding as many have noted that the r420 is having bilinear forced on it). The colored mipmap issue was already explained in the ATi chat and part of the PR misrepresentation. Whether or not the technical explanation is completely true depends on who you ask around here. :) I have seen shots between r420/rv350, r300, and 5900 that have shown mathematically identical results using trilinear versus adaptive filtering on here, at Rage3d, and at Beyond3D. Unless people start finding some pathological worse-case scenarios, I'd consider this a general case optimization of filtering that was horribly misnamed and marketed poorly.

[mikechai]

Well, after looking at those IQ comparisions again, and again... I actually think I changed my mind, certain parts look better on the nV card, such as the blending between mipmaps (trilnear filtering) While other parts look better on the ATI card, such as texture sharpness.. toss up for me.

Actually the transition of mipmaps is more noticeable and annoying during gameplay or in motion. That's what trilinear filtering is for.

[SmuvMoney]

Just how long have ati been cheating for? This is appalling. ati has acted like its superior and all this time its been duping everyone. But does anyone know when they started this stuff? Which driver build? I mean how long has stuff been going on for? :lame:

The adaptive filtering has been in the r9600 series since Catalyst 3.4 I believe. The 9700/9800 cards do not have this filtering. They are only capable of full trilinear, texture stage 0 trilinear, or bilinear.

[Bopple]

Looks like a flamebait...

The technique itself is not a cheating.
Actually, it's one of the best optimizations I have ever heard these days.

There is no such 'one and only absolutely right' method for filtering.
Full-trilinear filtering has been the best method, in most ways.
But it is not the holy golden standard in every way.
ATi's approach looks like a very good substitute for full-trilinear.

But the problem is, they claimed they were doing full-trilinear.
That's a lie. That's bad.
Perhaps some might claim that still ATi is better than nvidia. I agree.
But that doesn't negate the fact that ATi lied and it is bad.

All in all, I am not sure to label it a cheating. I'm inclined to label it a 'lying'.

One more: This is nvidia vga board.

[fivefeet8]

All in all, I am not sure to label it a cheating. I'm inclined to label it as a 'lying'.

When does the lie become a cheat? When benchmarks comparisons of "FULL Trilinear" are used by ATi's Request. -joking.

I'd also agree that's it's a "lying".

[Blacklash]

There are a few issues here,the main issue here isn't IQ but deliberate deception. Even telling reviewers how to handicap the 6800U vs their card in bench marks.

Maybe the adaptive trylinear does deliver close to the same quality in some cases, but we will never know because we are not offered the control over it to defeat it. Giving users that power would not harm folks that wanted to use/preferred it. The only potential harm could come from seeing the 16-20% loss on x800 performance numbers with full trilinear on every map, and that is the only reason you will never see it.

[neomoz]

Malficar got it one. :)

[Drumphil]

I still want a definition of "full trilinear"

[ChrisRay]

I still want a definition of "full trilinear"

Weighted Mip Map Blend...

[3madcows]

Every time I see more Q&A with ATI, it seems like they're digging themselves deeper into a hole. They still won't admit how often this technique is used in real games. Adding an option in the drivers to enable/disable would be their best response to the matter.

[Drumphil]

Weighted Mip Map Blend...

that describes what it does, not the technique is uses to achieve this. The term "Weighted Mip Map Blending"as far as I know would still cover ATIs technique.

bilinear filtering does weighted blending of a number of adjacent texels, trilinear does weighted blending bewtween adjacent texels and texels in the nearest mip map level.

So, does ATI's algorithm do blending between adjacent texels and mip map levels? As far as I can tell, yes.


They still won't admit how often this technique is used in real games.

100% of the time as far as I know. The question is wether or not its a bad thing.

[ChrisRay]

that describes what it does, not how its suppose to be done. Weighted Mip Map Blending as far as I know would still cover ATIs technique.



100% of the time as far as I know. The question is wether or not its a bad thing.


Trilinear filtering must sample 8 pixels and interpolate them before rendering, Doing twice as much sampling as Bilinear does. Anything less is not "true" trilinear filtering.

Nvidias Brillinear, and ATis Implementation is not "true" trilinear, its been defined as what I stated above, so calling it "trilinear" Filtering. When it is obviously not, Is Misleading.

[Drumphil]

where did you get that definiton of trilinear from?

And if it produces the desired result (blending between mip map levls), then does it matter what the algorithm is?

And if this is a "fake" version of trilinear filtering, lets see proof of the IQ loss it causes. Thats the real test.

[ChrisRay]

where did you get that definiton of trilinear from?

And if it produces the desired result (blending between mip map levls), then does it matter what the algorithm is?


Look up Trilinear Filtering on the web, That was off the top of my head, I'm sure you can find the same defintion however.

I'm not gonna answer your last question. Because its subjective, Mathematically, Its not doing the same thing however.

[jAkUp]

i think this is the most noticable part in the image... In the ATI screenshot you can clearly see where one mipmap ends. Overall the picture looks clearer on ATI though...

NV:
http://www.nvnews.net/vbulletin/attachment.php?attachmentid=5959
ATI:
http://www.nvnews.net/vbulletin/attachment.php?attachmentid=5960&stc=1

[Ruined]

Here are some more subtle examples of ATI doing less blending at mipmap transitions:

http://www.reflectonreality.com/nv17r420/cita.html
http://www.reflectonreality.com/nv17r420/citb.html
http://www.reflectonreality.com/nv17r420/

The idea of trilinear filtering is to provide a smoother transition between sharper and blurrier mips so you don't see lines between textures like the ones you can see in the ATI pics above. Of course that means that there will be an area of the texture that gets progressively blurrier as it approaches the mip, but that is simply so it transitions better into the blurrier textures.

[jbirney]

Trilinear filtering must sample 8 pixels and interpolate them before rendering, Doing twice as much sampling as Bilinear does. Anything less is not "true" trilinear filtering.

Nvidias Brillinear, and ATis Implementation is not "true" trilinear, its been defined as what I stated above, so calling it "trilinear" Filtering. When it is obviously not, Is Misleading.

So Chris, you know exactally what they are doing to get this? You know for a fact that they are not taking sample son 8 pixels and then you know how the handle those samples (ie they are not interpolate between them)?



Malficar Sig:

By default, one of our competitors does not offer Trilinear filtering, even when it’s requested by the application. By default, it uses an enhanced technique commonly referred to as “adaptive” filtering. This can not be disabled, after several mouse-clicks, from within their control panel.

What you failed to realize is that statement is 100% true.

[jimmyjames123]

i think this is the most noticable part in the image... In the ATI screenshot you can clearly see where one mipmap ends. Overall the picture looks clearer on ATI though...

You cannot judge the image quality here based on a still shot, as the mip map transitions would be evident during motion. Also, I believe that this pic you are looking at was actually captured using bilinear filtering on the ATI R420 card, and that's why you see such dramatic differences vs R360, NV40, and NV38 pics.