ATi is ch**t**g in Filtering

From what i gather from this thread, it seems that Ati's "fast trilinear" is very close (innocent until proven quilty) to full trilinear when it comes to IQ. It's still not "full" trilinear (which Ati already admitted) while marketed as such and that is the problem.

 

I agree.

 

IMU(npopular)O screw the users and the benchmarkers, but taking away the choice even from developers is the real annoyance here.

 

so is nvidias

http://www.beyond3d.com/forum/viewtopic.php?t=12486&postdays=0&postorder=asc&start=61

check daves comparison between 9800xt(trilinear)x800(brilinear)5950(brilinear)

 

I think the problem with Nvidias brilinear is that it's mostly visible in motion. This might of course be true of Ati's new method also.

But the problem is that there might be games where Nvidias brilinear is good enough compared to full trilinear. That's why i prefer to have it as an option (see Democoders post regarding reviews).

 

That might be true of ATI's method, but nobody has been able to find anything so far.
And there are loads of people in this thread trying their very best to find ANY loss of IQ in the ATI solution.

Those same people didn't have any problems finding IQ loss with Nvidia's solution. Most had immediately noticed it.

 

tEd, as I mentioned earlier in this thread, the position and textures of those images aren't necessarily the best for spotting such differences - I'm using them purely becuase they were easy to use.

It does have to be said though that NVIDIA's method has certainly improved from being perceptible or not over later drivers. The question remains whether those improvements are application specific or not (or, even if NVIDIA are able to disable this with presupplied mipmaps).

 

"Wild" guess: Andy Pomianowski = andypski?

 

Scrap em all and go with XGI!

Hey Dave...I think we should scrap them all and go with XGI or VIA!!!


*flashes Volari sign*

 

And where is the comparison and benchmark between Trilinear vs ATI's brilinear?
You just proved my point, that reviewers wouldn't treat it properly.
ATI makes a method that according to them doesn't cause any IQ loss compared to full trilinear. But you wouldn't even test that. Exactly what I predicted what would happen.

Basically what you are saying is that you don't have any way of objectively determining when IQ is good. Why do you need to compare to older cards? Why not a reference render?

If you are not able to determine if there is IQ loss whithout ATI giving you a full trilinear option to compare it to bit-for-bit. Doesn't that mean that they have done an excellent job, and that there is no problem?

It seems to me people are cyring foul play because of tiny differences that can only be detected when you do bit-for-bit comparisons, and are not detectable with the naked eye. Is that a good thing??
Aren't we forgetting that we are supposed to play games with this cards using are eyes instead of analyzing tools? (this applies to both ATI and Nvidia optimizations)

No!
The point is to deliver high quality and high performance to users!

Only when you cannot deliver both at the same time, and need to lower quality to get performance, THEN a user should be given a choice.

I still haven't seen a single person saying that ATI's solution gives lower IQ. When it is determined that ATI's solution gives lower IQ, THEN I agree that the user should be give a choice and that it should be an option for users.

But till then.... No harm done, so what is the problem?

 

I seem to remember that several months ago, Unwinder wrote an article about brilinear - what he found about it looking at the driver code.... he wrote that different mip-maps are compared and .... that's what I remember , but the article should be easy to find at nvworld.ru ..... (funny i had almost forgot about it....)

 

If you put the benchmarks in the same chart then you'd automatically get that comparision. For a similar situation, Ati's 2X FSAA was very good compared to Nvidias 4X on the NV3X. Under certain circumstances even better. But that doesn't mean that it would have been a valid comparision since the workload is completely different.

You could do a similar thing here. Review the different settings and mention that the quality of the "fast trilinear" is almost the same as "full trilinear".

 

IQ is so subjetive, color temperatures, saturation... there are more variables than pure filtering. Even fame influences judgement. As for the burden of de proof put on the consumers... it will take some time, but surely we will find out. But it should be the opposite.

I hope ATI put things straight tonight at the chat.

And finally, why not to give that option to users, reviewers, and developers so as to we can compare and say "Ahhhh, magnificent piece of work this adaptative filtering of ATI" all together?

Just my opinion

 

Don't know which one of you said this, but I like it enough to quote it.

Very well said in imho... 8)

 

I think someone needs to do a blind IQ test. Take a bunch of screenshots from different scenes in different games, with different IQ settings, resolutions etc, and package them up with an app that will pick two 'matching' images at random and ask the user to pick which he thinks is better. Then we could collect statistics that would let us figure out which settings can be viewed as having roughly the same IQ, and whether there actually is a noticable difference for these "optimisations".
Of course, the number of .png images required would make it a pretty big download, and it wouldn't help in cases where things are only noticable in motion, but it would be a step in the right direction, I think.

I seem to recall somebody doing a similar sort of survey for compressed audio formats a while back...

 

Don't you mean Magnificient piece of deceptive cheating? Of course, ATI conveniently tells reviewers to set the competitor's card to full trilinear, but forgets to tell that we won't be doing full trilinear in our cards.



ATI doesn't pratices what is preaches.

 

mjtdevries,

You are assuming as fact that image quality isn't compromised. You do not have a valid basis for that.

You might have reason to believe that this will be proven to be fact, but you can't jump ahead and disagree with people as if it was already proven.

...

Bjorn's example addresses how the hypothetical, as of yet, problem you see with DC's approach could be addressed suitably.

 

You must put this picture in your sig; i think you have posted it so many times i've lost count.

 

Yes, OK perhaps I didn't express myself correctly. What I meant to say is that games and other applications are written to an API that exists at the time the application is frozen. Even with OGL, even the most forward-looking application can only reliably be written to extensions that exist at the time of writing. When NV_vertex_shader(N+1) comes along, the game knows nothing about it.

If filtering, etc., was exposed in the same way as pixel/vertex shaders, older titles wouldn't be able to benefit from the apparently very rapid *ahem* advances *cough* in image quality that the drivers from the IHVs are giving us.

So I guess what I meant was that game/apps represent the state of the API at the time the game/app was frozen.

 

Ahh but you see, with their "trick" performance goes up and its desirable from a quality standpoint.