Ways to solve complaining about ATI's filtering?

[Mintmaster]

love for someone with Mafia and a 9600 to test that road image with bi and try, and tri and bri. Ante P must have the two around. Let's see if he reads this.

Ooooh! That'd be a great game to compare it on, the road really would give it away.

I'll hunt around for me Mafia disks, it's been ages since I played it anyways and it's a great game!

I'm looking forward to the results, dig. Road lines are the best way to see the effects of filtering in an actual game, but they have to be low res like in the older NFS (HP and PU) games. As you watch them come from a distance, they would sharpen in steps rather than completely smoothly. Racing games, IMO, have one of the biggest needs for anisotropic filtering.

Quitch, this is one example where the difference between brilinear filtering and trilinear filtering is noticeable, and could be distracting.

Having said that, however, once you enable anisotropic filtering it's hard to spot any differences.

 

[Sandman]

I have to wonder (out loud, of course ) about whether or not this test is helping to show that ATI's optimization is perfectly valid.

As of now, only 38% of the respondants correctly identified true trilinear on a pathological worst case. Also, remember that B3D members really know their stuff, so I wouldn't expect any other group to do better in a similar poll (assuming no random guessing). Combine these results with the in-game screenshots from another thread, where only a couple of people claimed to see a difference, and you have a strong case that the IQ is pretty much unchanged.

disclaimer: Of course, seeing the optimization in a game might show things that static shots can't reveal.

 

[Evildeus]

If i can spot the true trilinear it's because one is doing a better job, hence the other less quality in doing this particular job. So tell me to read all you want that won't change a thing cause you can't say anything else than READ, so be quiet, stop trying to be almight and stop yelling.

PS: if you are sick go and get your pill ok?

quality - The totality of features and characteristics of a product or service that bear on its ability to satisfy stated or implied needs. Not to be mistaken for "degree of excellence" or "fitness for use" which meet only part of the definition.

http://computing-dictionary.thefreedictionary.com/Quality

 

[Rolf N]

Actually it does, because it will begin to use lower mip levels earlier.

No and yes. It begins to use lower resolution mipmaps earlier than bilinear does its "switch" thing. It will also keep higher resolution mipmaps "around" longer. It's important to note that mipmap transitions are not in the same spot. Bilinear switches to a new mipmap level roughly in the same spot where trilinear reaches a 1:1 blend.

Imagery on a Radeon 9500Pro, Q3A in 640x480.
colored:

result:

If you look towards the transition from the first (uncolored) to the second (red) mip level, IMO you'll see that bilinear is indeed blurrier there.
You'll be hard pressed to see much more difference between trilinear and bilinear in this example, but I'd still say that bilinear is never sharper.

 

[WaltC]

I have to wonder (out loud, of course ) about whether or not this test is helping to show that ATI's optimization is perfectly valid.

As of now, only 38% of the respondants correctly identified true trilinear on a pathological worst case. Also, remember that B3D members really know their stuff, so I wouldn't expect any other group to do better in a similar poll (assuming no random guessing). Combine these results with the in-game screenshots from another thread, where only a couple of people claimed to see a difference, and you have a strong case that the IQ is pretty much unchanged.

disclaimer: Of course, seeing the optimization in a game might show things that static shots can't reveal.

It's the "eyewitness" syndrome, coupled with the fact that there were only two real choices available--right & left. A random guess has a 50% probability of being correct in such a case, even if you were blindfolded... I'd like to see them try again with maybe six or eight choices, with bilinear thrown in, as well as some FSAA and AF, and I very much doubt anywhere near even 38% would be able to correctly sort the optimized trilinear choice. I couldn't convince myself one way or the other after staring at it for maybe five minutes and looking at it from different angles.

I think what troubles me about all of this is that it is the inverse of the original nVidia brilinear situation in which it was visible mipmap boundaries in games like UT2K3 that brought it to people's attention. In this case, it seems that ATi's admission it is optimizing for trilinear, as opposed to plainly visible detriments in IQ, is the "catalyst" for the criticism, so to speak... That strikes me as a waste of time.

My position is simply that if the mipmap boundaries disappear, then I don't care what they call the filtering method, or how they do it. Whether it's nVidia or ATi, makes no difference. The only reasonable and sane objection to have about a trilinear optimization, in my view, is that it doesn't do as good a job at masking mipmap boundaries as is possible with standard trilinear. Barring that, I see no other reason to object.

 

[noko]

Interesting, it took me less then 5 sec to know which was the real Tri. In motion I am sure I could have notice it instantly.

 

[croc_mak]

Interesting, it took me less then 5 sec to know which was the real Tri. In motion I am sure I could have notice it instantly.

Interesting...So, do you have a X800 or a 9600? If so, did you notice anything instantly that you would like share?

 

[Althornin]

If i can spot the true trilinear it's because one is doing a better job, hence the other less quality in doing this particular job. So tell me to read all you want that won't change a thing cause you can't say anything else than READ, so be quiet, stop trying to be almight and stop yelling.

PS: if you are sick go and get your pill ok?

Evil, i am sick of explaining this to you.

There is a difference between being able to distinguish a DIFFERENCE and determining which one is BETTER.
You are apparently unable to understand that, so i am telling you over and over again, and you are failing to comprehend.

Maybe you can open your eyes and realize WHY DaveB is doing both of his new polls - to answer the question you THINK the first poll did.

And i'll keep telling you to read, until you do, and understand.

If you can spot the trilinear, it is DOING SOMETHING DIFFERENT.
DIFFERENT DOES NOT EQUAL BETTER.
Get that through your head.
If you cna spot the trilinear, it is because you know to expect to see the mip transistions at a different spot with it. You can see the transitions on BOTH of them. Why is this like talking to a brick wall? again, DIFFERENT DOES NOT EQUAL BETTER.

For instance, you might look at a blue shirt and a red shirt, and tell yourself - "Hey, i can tell the difference between these!" - but that doesnt make the red shirt the best one, simply because you can tell it from the blue.

No matter how much you want the previous poll to mean what you say it does, it DOES NOT.

So maybe you can be quiet, and take a pill of your own - one that aids comprehension.

note: I am not saying trilinear is NOT better. I am not saying ATI's new filtering is equal to trilinear in quality. What i am saying, have been saying (if you'd bother to read) and will continue to say until you show signs of comprehension, is that the poll you keep referencing does not shwo what you claim it to, no matter how much you WISH it did. Wait for the result of DaveB's new polls, and then open your mouth about something relevant.

 

[bloodbob]

If you can spot the trilinear, it is DOING SOMETHING DIFFERENT.
DIFFERENT DOES NOT EQUAL BETTER.
Get that through your head.

Yes because its different doesn't mean its better doesn't mean its worse,

Though I think if I come with some new brillant maths and I don't give out a proof that it is correct that we should ultimately assume it is better and correct.

This is what ATI is asking us to do they are saying their new formula is BETTER without proof and we are just supposed okay its better until someone comes along and disproves it?

In my eyes that stupid and I know some users on the forum are telling me I should be accepting its better until I can disprove them. Without and proof or any verification even. I think you should ultimately be conservative.

 

[Evildeus]

Evil, i am sick of explaining this to you.

Take your pill ok? If you need to say 500000 times the same thing, it doesn't make it right.

There is a difference between being able to distinguish a DIFFERENCE and determining which one is BETTER.

The fact that there's is a difference means the quality is not the same, otherwise there's no difference in making the same thing. If it was a different thing (like your exemple), then i would agree. Sure it doesn't mean better it could be worse.

You are apparently unable to understand that, so i am telling you over and over again, and you are failing to comprehend.

You are the one that don't understand. I completly understand what you are saying, and i disagree. If your application is supposed to do b and you do b-, there's a difference. If there's a difference, one is making a better job than another. I don't think it's difficult to spot.

Maybe you can open your eyes and realize WHY DaveB is doing both of his new polls - to answer the question you THINK the first poll did.

Maybe the 2 new polls are worse than the precedent?

What is the R420 pic? Heck how i would know? I don't have one, i don't know what are the others, i don't know what are the filtering etc.

Which one does look better? In doing what? Globally or in particuliar effect? etc. I have said multiple times, that i'm not talking of globality but of one particular effect. Look at what ingenu wrote.

And i'll keep telling you to read, until you do, and understand.

I've read and understood. Now try the same thing for once.

If you can spot the trilinear, it is DOING SOMETHING DIFFERENT.
DIFFERENT DOES NOT EQUAL BETTER.
Get that through your head.

so if i can spot the trilinear, it's because it's doing something different. Different from what point of view? From what trilinear is supposed to do? If it's different from what trilinear is supposed to do then the difference is on this particular aspect. right? In this particular aspect, one is doing the job, trilinear, the other is doing another job, which is different. If the job is different, the characteristics the feature that bear on its ability to satisfy stated or implied needs its ability to satisfy stated or implied needs are different (definition of quality) hence a difference of quality :?

If you cna spot the trilinear, it is because you know to expect to see the mip transistions at a different spot with it. You can see the transitions on BOTH of them. Why is this like talking to a brick wall? again, DIFFERENT DOES NOT EQUAL BETTER.

Thanks, you just proven my point. Yes i know what to expect, yes i can spot the transition in both, i never said the contrary. Trilinear is the standard in the case we are talking. I can see more transition in the R420 situation, hence less quality in doing this particular job. I never said that overall the image quality was better with the trilinear one.

For instance, you might look at a blue shirt and a red shirt, and tell yourself - "Hey, i can tell the difference between these!" - but that doesnt make the red shirt the best one, simply because you can tell it from the blue.

No, that's not a good exemple, and i understand now, why i can't agree with you. A better exemple would be to do a certain blue for a t-shirt, hence which one is doing a better job in doing so. But there's an image on the t-shirt, and even if one is doing a better job in doing the blue, it could do a worse job in overall because the image is really worse.

 

[Pete]

You'll be hard pressed to see much more difference between trilinear and bilinear in this example, but I'd still say that bilinear is never sharper.

"Never" is such a strong word. I always assumed bi was most noticably sharper than tri because at the end of the first mip level, tri is slightly blurrier beacuse it begins to blend with a smaller image. I never knew that tri ended up slightly sharper at the beginning of the second mip level, though, as your pic shows. At least, I'm fairly convinced--both from just eyeballing it and just thinking about it--that it's really sharper, and not just a false perception (read: visual trick) caused by how blurry the second mip level looks compared to the first with bi's stark transitions.

For instance, you might look at a blue shirt and a red shirt, and tell yourself - "Hey, i can tell the difference between these!" - but that doesnt make the red shirt the best one, simply because you can tell it from the blue.

No, that's not a good exemple, and i understand now, why i can't agree with you. A better exemple would be to do a certain blue for a t-shirt, hence which one is doing a better job in doing so. But there's an image on the t-shirt, and even if one is doing a better job in doing the blue, it could do a worse job in overall because the image is really worse.

Well, this is the crux of the issue: are Dave's polls asking which shot looks better (red/blue shirt), or which looks most like true/legacy trilinear (blue/blue) shirt? If we've decided that true/legacy trilinear is the best filtering method, then I suppose it's more useful to approach Dave's polls as blue/blue ones (i.e., objective, or "Which gets closer to a known target?"). But I'm approaching them as red/blue ones (i.e., subjective, "Which do you prefer?"). As Dave

1) went out of his way to NOT mention trilinear,
2) actually named one of the polls "Which is your preference?", and
3) posted two polls for the same set of pics,

I conclude that he's aiming for quality, not legacy. IMO, he's looking for red/blue, not blue/blue. I don't know enough about polling to determine if the quality and combination of the two poll questions is enough to eliminate inherent biases (tri = best and ATi = evil).

Or maybe I'm overanalyzing the situation, or just plain talking out of my butt. Both are distinct possibilities.

 

[Evildeus]

No you are right on but i was speaking of the first poll.

 

[Pete]

Sorry, my fault for skimming both your post and the forums (I somehow both forgot about and didn't see the first poll--bah).

 

[Sandman]

The fact that there's is a difference means the quality is not the same, otherwise there's no difference in making the same thing.

I have to agree with Althornin, you can have differences in 2 things that are not related to quality. As a quick example, let's assume that ATI's hidden surface removal algorithm is much better than nVidia's. Thus, an ATI card will do less work than an NV card to render the exact same image. In this situation, you'd have a difference in making the same image without any difference in quality.

 

[croc_mak]

This is what ATI is asking us to do they are saying their new formula is BETTER without proof and we are just supposed okay its better until someone comes along and disproves it?

In my eyes that stupid and I know some users on the forum are telling me I should be accepting its better until I can disprove them. Without and proof or any verification even. I think you should ultimately be conservative.

What I read ATI saying is that there method is NOT demonstrably better or worse than regular trilinear and is only marginally faster. I see there is no need to announce something like this other than driver relase xyz is 5-10% faster in abc applications...

Don't we have enough buzzwords already?

 

[Evildeus]

The fact that there's is a difference means the quality is not the same, otherwise there's no difference in making the same thing.

I have to agree with Althornin, you can have differences in 2 things that are not related to quality. As a quick example, let's assume that ATI's hidden surface removal algorithm is much better than nVidia's. Thus, an ATI card will do less work than an NV card to render the exact same image. In this situation, you'd have a difference in making the same image without any difference in quality.

Exept we are effectively talking about the image, not how to render it. anyway, Ati would do a better job, so their removal algorith is higher quality (hence better performance) than Nv's one.

 

[bloodbob]

This is what ATI is asking us to do they are saying their new formula is BETTER without proof and we are just supposed okay its better until someone comes along and disproves it?

In my eyes that stupid and I know some users on the forum are telling me I should be accepting its better until I can disprove them. Without and proof or any verification even. I think you should ultimately be conservative.

What I read ATI saying is that there method is NOT demonstrably better or worse than regular trilinear and is only marginally faster. I see there is no need to announce something like this other than driver relase xyz is 5-10% faster in abc applications...

Don't we have enough buzzwords already?

Go read what they said again

Andy/Raja
We have never removed "full trilinear". We certainly do not believe that any benchmarks have been invalidated by our techniques. In all cases reviewed so far we believe that we have higher image quality than other implementations.

So that means every other so called trilinear including "traditional" trilinear has lower image quality.

 

[noko]

Interesting...So, do you have a X800 or a 9600? If so, did you notice anything instantly that you would like share?

Na, wished I had one just to experiment with. Looks like I will be picking up a 6800 sometime in the future. My biggest concern would be obvious mipmap lines in motion. Now if more testing was actually done on a variety of games and a variety of conditions the better I would understand if IQ was degraded significantly or not, if at all. But a mute point it has become since I will probably be getting a 6800 anyways.

 

[KimB]

I think what troubles me about all of this is that it is the inverse of the original nVidia brilinear situation in which it was visible mipmap boundaries in games like UT2K3 that brought it to people's attention. In this case, it seems that ATi's admission it is optimizing for trilinear, as opposed to plainly visible detriments in IQ, is the "catalyst" for the criticism, so to speak... That strikes me as a waste of time.

Um, no. The thing that sparked this whole controversy was the difference in benchmark scores between colored MIP maps and normal.

The only reasonable and sane objection to have about a trilinear optimization, in my view, is that it doesn't do as good a job at masking mipmap boundaries as is possible with standard trilinear. Barring that, I see no other reason to object.

Since we don't know what optimization ATI is implementing, we cannot know if there are other effects, effects other than the traditional bilinear filtering MIP map boundaries.