Examples of ATI filtering failing?

Yes and No (from how I understood the results). The cases shown in the IQ Poll show what trylinear would look like if it would be enabled on pathological worst-case basis. However, the drivers/hardware disable trylinear in those cases and fall-back to trilinear. So the user never sees trylinear looking that bad.

As for your point 1, you do not even have mathematical equals for trilinear between different implementations (video card manufacturers and reference rasterizers).

 

Well I doubt that is the worse case but he said has anyone got an examples of where its failing. It has failed here simple as that.

Actually thats is slightly true because everyone up till now has been taking a very small approximation with the LOD calculation I believe. This is where the NV40 is doing it right . I think you will find their is a mathematically equal between the R3XX and the reference rasterizer I think you will find that ATI not only uses the same calculation they use exactly the same precision for the linear blend.

 

3DC user grestorn claims to have filtering issues in Far Cry and Deus Ex 1. He's got an X800 Pro. Link (German).

[translation]
"Trylinear" is visible. That doesn't mean it's inacceptable. E.g. in Far Cry there are apparent vertical waves on interior walls, with strong texture shimmering. In Deus Ex 1 there are harsh mip map transitions. Both of these issues don't exist on the gf's Radeon 9700.
[/translation]

edit:
According to Computerbase (see ram's posting below), R420 filtering differs from RV350's by a rather stupid LOD tweak. This would explain the shimmering in Far Cry on R420, and also means that RV350 is likely not affected.

 

(Before I comment on this let me say that I'm not in the position to evaluate ATI's "optimization". It might be good or bad, I've not seen it in real life yet myself.)

I think you got it wrong. As far as I understand it, ATI does not claim that their "optimized" trilinear looks equal or better than real trilinear all the time. They claim that their "optimized" trilinear looks equal or better in specific cases where the mipmap levels are just simple downsamples of the real texture. And consequently they only enable their optimization when the driver detects that this specific case is given.

 

ComputerBase (german) has some examples where the detection algorithm fails where it is obviously that trilinear would be needed.

 

i was hoping for a b3d article on the changes that happened to tri and af on the x800, back in the days it would have happened more likely imo.

 

You may have noticed that we haven't been lacking in things to write about recently - I've hardly had any time to myself since April. If people are willing to pay me to do this full time then I will and then I'll have more time to look at more things, until then I only have a finite quantity of time and you'll have to accept what we give.

You'll also note that the majority of articles looking at this so far have taken it from a theoretical level and having been coding tools to specifically highlight the issue. For one, we don't have the coding resource to look at it (however, we generally use what people give us, hint hint) and second, as this thread bears evidence to, there are still very few actual examples of notable differences in games. I am, however, still waiting for an interview to come back from ATI (I have been chasing them) on the subject and if I get some time I might take a closer look look at some game differences then.

 

Dave as allways all your articles are top notch and I've read the newest ones a few times .

Take your time on this and when its done its done . As for money throw more adds up . I am allways suprised at the lack of adds on this forum and the main page its self.


And if you need help well I know nothing compared to alot here but if its just benhmarking or taking screen shots from cards I have acess to well I can help out as much as possible.

 

hi dave,
i can understand the time issue completely. what i dont understand is the mixing of topics. for a pure gamers site the 'gaming relevance' argument holds water but on b3d?

 

Again, I can use the tools at my disposal to use - those tools are either are / in games or coded elsewhere. If none of the tools at my disposal highlight any issues to any great extent what am I supposed to do?

You may remember in in initial thread it was us that highlighted that this had occured since the first set of drivers for 9600.

 

I'd like to add something that most of you seem to be missing (or forgetting): what prevents ATI's algorithm from improving? Nothing; ATI has specifically stated that if in some cases the algorithm fails, they will remedy it. So if/when we find occassions that the algorithm fails within games, and ATI fixes that "glitch", wouldn't that be better?

 

The question is then, is it a "general" algorithm or do they have to add app detection for certain games ?

And, should we or should we not use "force trilinear" for Nvidia when doing "head to head" reviews ? (Dave has hinted that "brilinear" is much better now then it used to be)

 

I am preparing an editorial about all that, but I will tell you this. For starters, app detection for these sort of things is a no-go for me, there are ways to keep your algorithm generic by programming for certain situations, not applications. ATI says it's general, and I have no reason to doubt them.

Now to your next question, I believe that we should focus on the end result rather than the means to get there. What I mean is this: nVidia produced an algorithm that was not efficient visually. Yes it was faster than true trilinear. So if that is how nVidia wants to be benched, let her have it. Bench with brilinear on. ATI wants to be compared by using their own algorithm? Let them be compared using that! When a person without much knowledge of the whole 3d gfx reads the review, you know what he should read? "We compared both cards, and although nVidia was a little faster, its visual quality was much worse than ATIs". Let the buyer make the decision what he wants to buy. We are here to present the facts to him, not make the decision FOR him.

There are people out there who swear by FPS, and people who swear by IQ. You can't please them all. All you can do is point what card will suit their particular needs. True journalism is about presenting the facts and let the reader decide, not presenting your view on things and project your opinion on him.

 

Surely if it WERE dynamic then it WOULD disable any attempt to optimise when colour mipmaps were on because the transitions are so extreme. This is an example of the thing working. I really don't know what you're talking about, and frankly I thought we were past this stage.

 

I'm sure it is now. But it seems that i has problems in certain games (i guess we need more info on this though) and what if the current algorithm can't handle this ? If Ati can't tweak the algorithm to handle this then will they return to "normal" trilinear ?

I agree. But let's forget all about equal workload then (contrary to what Ati suggests in their .pdf).

Is "brilinear" really that much worse then ? If it has improved a lot lately, what does it actually look like with f.e 16X AF, in, let's say Far Cry f.e ?
(i'm not stating that it looks good, i'm asking because i don't know)

True. But the problem in this business is rather often, is the facts really the facts ? And do i agree with your opinion of the facts ? (as in trylinear = trilinear, brilinear = tryliner .....)

 

The way i see it you can compare the output of two cards, and the result could be a) Card A has better IQ than card B b) Card B has better IQ than card A c) both cards have roughly the same IQ. The "bad" thing about image quality is that it is purely subjective. Some people love Digital Vibrance's way of altering the image's colours, some people hate just that. What is "right" and what is "wrong" in that occassion?

That is the reason why I am saying that we should stick to the facts as far as IQ tests are concerned. If the reader has the facts in his hands, he will make the decision according to his tastes and beliefs. If ATI or nVidia want their products to have bad IQ, let them. Show the reader what he'll get for his money and let HIM decide whether he wants to spend his money on that.

P.S the "ATI/nVidia" example i used in my earlier post was purely speculation, nothing else; i just made it to show my point more clearly.

 

What are the facts that you should present to the user then ? cause simple screenshots won't do it when it comes to filtering. And Mip Map coloring tools is not enough either

 

Hence we need to sit down and discuss things, all major reviewers. My whole point exactly.

 

So if these don't do it we should subtract images from other images to see a diffrence ? that is a bit nuts .

Why hasn't anyone posted a video of problems they are seeing with the x800s and 9600s ?

I think we all know that they answer to that question is no one is seeing problems .