ATi is ch**t**g in Filtering

[davepermen]

No. I'm not saying that at all. I'm saying that it will happen by accident. In other words, I don't be believe that ATI's solution could possibly handle all corner cases.

give any reason to believe you. one hint: if it is, by any chance, just in one case, a corner case, ati falls back to trilinear. guess what? coloured mipmaps ARE such a corner case, and that is WHY they do fallback.

give me any believe there is a corner case.

and give me any reason to see why this should be called a cheat. it is trilinear where trilinear is needed. it is quality based filtering where quality based filtering is what is needed. it is bilinear only where bilinear is what is needed. etc..

do you program? i'm curious. as a programmer, knowing the side of the graphics-developers, i can certainly say, in my point of view, NO customer should complain this time. brilinear was another issue. quality degradation IS an issue. this is not quality degradation. not by any chance (except bug in driver, wich is unlikely.. because the bug-catcher simply does trilinear.. )

 

[martrox]

Guys, give Chalnoth a break here.... do any of you believe a tiger can change it's stripes? IF you expect his nVidiacentric logic to make any sense, the the jokes on you..... it seldom has and seldom will!

And the proof of this is....well, just look at all the complaints about this on the RV350/360 cards.........

Sometimes, don't you wish there was an "ignore" button?

 

[Deleted member 2197]

Good points regarding ATI's deception of reviewers, and the general public.

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

 

[martrox]

Good points regarding ATI's deception of reviewers, and the general public.

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

Not really...and it's already been discussed here.........

 

[mikechai]

and give me any reason to see why this should be called a cheat. it is trilinear where trilinear is needed. it is quality based filtering where quality based filtering is what is needed. it is bilinear only where bilinear is what is needed. etc..

Because they claimed they are doing full trilinear and in fact, they are not(if like you said, they are doing trilinear only in situation where they think is needed), therefore its a cheat/lie(whatever you call it).

If you'd seen the ATI benchmarking guidelines sent to reviewers, ATI set the rules, they want reviewers to use full trilinear on their card and make sure that their competitor uses full trilinear all the time. But then they themselves not running full trilinear even when it was set in CP and applications. Instead, they use an optimized algorithm to only do trilinear when needed and this reduces workload and therefore has a higher fps than they would have if running full trilinear.

To make things worse, they recommend reviewers to use colored mipmap to check filtering quality which in fact showed that they are doing full trilinear, while reviewer didn't realise that that is the only case that the card will do full trilinear.

As I know, some of the reviewers were given only a short time to do the preview. And in timedemos, I think it is impossible to notice anything wrong because the frames are flying so fast. Furthermore, reviewers tend to rely/trust on colored mipmap to check filtering quality.

I have no problem that ATI introduces a great method of smart filtering. I think its a good trade off because it gives a comparable quality at faster performance. But please don't deceive the public and the press that they are doing "full trilinear" filtering all the time. Whatever they have done will make us think that they desperately want a more favorable benchmark numbers.
________
SUZUKI SV650

 

[Mariner]

To make things worse, they recommend reviewers to use colored mipmap to check filtering quality which in fact showed that they are doing full trilinear, while reviewer didn't realise that that is the only case that the card will do full trilinear.

I think they've actually said it's not the only time they will be doing full trilinear. It is just a particular case where full trilinear is necessary and the card should perform full trilinear in any case where the algorithm deems it necessary.

I don't think ATI have handled the whole situation as well as they perhaps could have but to me the fact that nobody seems to have complained about the filtering quality of the RV3X0 series is very relevant. These cards have been around for some time yet nobody has complained about poor filtering quality - this indicates to me that the algorithm must work pretty well.

I'm not quite sure what to make of NVidia's continued silence on the matter. On the one hand I'm pleased that they aren't mud-slinging as they have been known to do in the past, but I can't help but wonder if it is because they, themselves, expect their trilinear optimisations to be widely used.

 

[Dave Baumann]

To make things worse, they recommend reviewers to use colored mipmap to check filtering quality which in fact showed that they are doing full trilinear, while reviewer didn't realise that that is the only case that the card will do full trilinear.

You might notice they also recomend people use comparison tools to check the outputs - they were actively encouraging people to look for differences (which may actually suggest an oversight of this being there).

 

[Deleted member 2197]

Good points regarding ATI's deception of reviewers, and the general public.

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

Not really...and it's already been discussed here.........

Perhaps, but this is the first article I've seen where ATI's deception is not overlooked with the promise of better IQ, in other words no attempt is made to "sugar coat" what was done. A cheat is a cheat ...

 

[davepermen]

Because they claimed they are doing full trilinear and in fact, they are not(if like you said, they are doing trilinear only in situation where they think is needed), therefore its a cheat/lie(whatever you call it).

they ARE DOING FULL TRILINEAR WHERE IT IS REQUESTED. it can not be requested by the user, but by the application developers. if i, as gamedev, want a full trilinear filtered texture, i get that. 100% all the time on any ati hardware. this was/is NOT the case on nvidia hw with brilinear. and THIS is where the cheating happens.



some people .....

 

[lowdog]

You might notice they also recomend people use comparison tools to check the outputs - they were actively encouraging people to look for differences (which may actually suggest an oversight of this being there).

Pride comes before a fall.

 

[Martillo1]

Edit: answering to mariner

Surely they will have ATI fanboys blessing it

And I expect to see shader optimizations vía mathematic transformations as well. And better no one complaints about it if the output look almost equal.

As it is said in spanish: "o jugamos todos o se rompe la baraja"

 

[mikechai]

I don't think ATI have handled the whole situation as well as they perhaps could have but to me the fact that nobody seems to have complained about the filtering quality of the RV3X0 series is very relevant. These cards have been around for some time yet nobody has complained about poor filtering quality - this indicates to me that the algorithm must work pretty well.

I didn't say their algorithm gives poor filtering quality, in fact it works well. Although I think *traditional* full trilinear will be better(but worse performance).

But then RV3x0 users have not experienced full trilinear yet on their cards. As you know the option is not available to them. Therefore, they can't compare.

I still think the option should be there for users/reviewers to decide what they prefer.
________
Dodge 100 \commando\"" history

 

[mikechai]

they ARE DOING FULL TRILINEAR WHERE IT IS REQUESTED. it can not be requested by the user, but by the application developers. if i, as gamedev, want a full trilinear filtered texture, i get that. 100% all the time on any ati hardware. this was/is NOT the case on nvidia hw with brilinear. and THIS is where the cheating happens.



some people .....

It is not. (that's the problem)
________
FORD ESSEX V6 ENGINE (CANADIAN)

 

[davepermen]

they ARE DOING FULL TRILINEAR WHERE IT IS REQUESTED. it can not be requested by the user, but by the application developers. if i, as gamedev, want a full trilinear filtered texture, i get that. 100% all the time on any ati hardware. this was/is NOT the case on nvidia hw with brilinear. and THIS is where the cheating happens.



some people .....

It is not. (that's the problem)

it is.

 

[mikechai]

I'm out of this thread.
________
HAWAII MARIJUANA DISPENSARIES

 

[tEd]

they ARE DOING FULL TRILINEAR WHERE IT IS REQUESTED. it can not be requested by the user, but by the application developers. if i, as gamedev, want a full trilinear filtered texture, i get that. 100% all the time on any ati hardware. this was/is NOT the case on nvidia hw with brilinear. and THIS is where the cheating happens.



some people .....

It is not. (that's the problem)

it is.

on r3x0 yes not on rv3x0,r420

 

[bloodbob]

Because they claimed they are doing full trilinear and in fact, they are not(if like you said, they are doing trilinear only in situation where they think is needed), therefore its a cheat/lie(whatever you call it).

they ARE DOING FULL TRILINEAR WHERE IT IS REQUESTED. it can not be requested by the user, but by the application developers. if i, as gamedev, want a full trilinear filtered texture, i get that. 100% all the time on any ati hardware. this was/is NOT the case on nvidia hw with brilinear. and THIS is where the cheating happens.



some people .....

Ehh Davepermen as a dev how I can force ATI to give me trilinear without A) generating the mipmap by myself rather then letting them become autogenerated. B) being forced to use another minification filter for generation of the mipmaps??? 90% of textures use auto-generated mipmaps with the box filter which ATI refuses to do trilinear with. Its even worse trying to do it in OpenGl.

 

[davepermen]

do the mipmaps on your own.. if you use boxfiltering, you get the nice optimized thing.. so what? if you don't box-filter, you won't get it. (you shouldn't..)

if you enable automatic mipmap generation, you have no choise anyways. i liked to use it on rendertexture tests, but guess what? it differed depending on card (gf2 days) that much that it wasn't usable.

it isn't much to do it in opengl. and if you read the opengl spec, it's all valid and allowed how they do it. they are, indeed, allowed to filter with what ever algo they want, if it at least gives the expected result. how they do the trilinear filtering doesn't mather. it's called abstraction.

oh, and, ati and nvidia both cheat to us! if they upload a texture and realise it doesn't really need 32bit colours but can work with 16bit colours, they reduce it, to save gpu-memory, and bandwith! CHEATING ALARM!!!!




some....

 

[Dave Baumann]

do the mipmaps on your own.. if you use boxfiltering, you get the nice optimized thing.. so what? if you don't box-filter, you won't get it. (you shouldn't..)

Are there any rules for Box filtering?

 

[bloodbob]

do the mipmaps on your own.. if you use boxfiltering, you get the nice optimized thing.. so what? if you don't box-filter, you won't get it. (you shouldn't..)

if you enable automatic mipmap generation, you have no choise anyways. i liked to use it on rendertexture tests, but guess what? it differed depending on card (gf2 days) that much that it wasn't usable.

it isn't much to do it in opengl. and if you read the opengl spec, it's all valid and allowed how they do it. they are, indeed, allowed to filter with what ever algo they want, if it at least gives the expected result. how they do the trilinear filtering doesn't mather. it's called abstraction.

oh, and, ati and nvidia both cheat to us! if they upload a texture and realise it doesn't really need 32bit colours but can work with 16bit colours, they reduce it, to save gpu-memory, and bandwith! CHEATING ALARM!!!!

Box filtering is best preformance/quality combo ( possible the best quality I dunno if it is considered better then tent/triangle filtering or not but it is certainly quicker ) so if I wanna use it I should be forced by IHV to change my program. And why should every developer be forced to write their own code for mip-map generation isn't one of the whole points of DX to make things easier for the developer. If your a program then you know you should reuse code as much as possible so forcing each ISV to write their own mip-map generation code is stupid.