ATI cheating on anisotropic filtering performance

[PatrickL]

The sad thing is we know Chalnoth posted that as a provocation but that Richtofen is serious

 

[Tim]

bottom line is this:

in directx they optimize on filter stages.

Yes, and?

It is not easy to turn those optimizations off unfortunatly 3dcenter.de
were able to do that and guess what ... there was a performance hit on the R420.

If the game request trilinear it gets trilinear.

Most reviewers overlocked that fact. They just benchmarked and did not analyse this matter.

Most reviewers benchmarked with briliniar filtering on nVidia cards too, guess what ... there was a performance hit on the NV40 when it is turned on.

Give me just 2 minutes and i will find thousands of threads and postings on various forums including this one here where people accused Nvidia of cheating while doing something like that.

You will fail what people was complaining about was that nVidia cards didn’t do what the games requested not that they offered optimizations.

Where are all those people right now?

They have never existed the nVidia issue was completely different.

AT 3dcenters request ATI responded that they will continue to use that optimization.

Why not.

Nice one... really nice one and espacially really nice how different the vast majority is viewing at both companies in this matter. The one is cheating the other one just optimizing.

It is very simple you don't cheat if you do what the game request.

My standpoint is clear. If Nvidia was cheating while using lower precision or using a stage optimization of FX cards ATI this time is simply flat out cheating too.

No.

That is the bottom line and there is no way to deny that.

Only when you ignore the facts.

 

[LeGreg]

So how do we know when we read a review if the editor
was using control panel aniso or app aniso ?

I obviously don't talk about b3d's review since they don't
do head to head so it's a useless point in that case

 

[Dave Baumann]

I obviously don't talk about b3d's review since they don't
do head to head so it's a useless point in that case

Besides, as I said, its listed in the Test Setup page.

 

[anaqer]

One tiny question more : what's the big deal after all? Does ANYBODY actually use the CP to set AA / AF?

Get aTuner.
Use it.
End of story.

 

[Moose]

So how do we know when we read a review if the editor
was using control panel aniso or app aniso ?

I obviously don't talk about b3d's review since they don't
do head to head so it's a useless point in that case

If you look at the 6800 review and the x800 review here at B3d you will see that they were benched and tested in exactly the same way, so the results are comparable.

As far as what other sites do, unless they say how they are testing then you won't know. Do the sites say if they have Nvidia's AF optimizations turned on or off???

If the site doesn't say what settings they used etc then it isn't a very good review and shouldn't be trusted at all.

 

[Neeyik]

One tiny question more : what's the big deal after all? Does ANYBODY actually use the CP to set AA / AF?

Get aTuner.
Use it.
End of story.

I do. Dead easy - right click on the system icon (NVIDIA card user), select what you need, run game. Why bother with another app when the functions I need are already there?

 

[Ailuros]

One tiny question more : what's the big deal after all? Does ANYBODY actually use the CP to set AA / AF?

Get aTuner.
Use it.
End of story.

RadLinker for moi

 

[anaqer]

Why bother with another app when the functions I need are already there?

I hate everything that dares enter my systray & run in the background with a passion... unless it absolutely, positively has to. ( Hell, I used to disable the volume dial before TV forced me to use it regularly. )
Changing AA / AF is something you do when a new game comes along that makes you crank it down temporarily, then you set it back... don't know about you, but to me that doesn't exactly warrant another systray icon, another process, another few megs... :?

 

[Randell]

I hate everything that dares enter my systray & run in the background with a passion...

that doesn't exactly warrant another systray icon, another process, another few megs... :?

Radlinker does non of those things

 

[andypski]

bottom line is this:
in directx they optimize on filter stages.

No we do not - in DirectX we do what the application requests. If you choose to go outside of DirectX specifications by using the control panel (please find me the specifications for that in the DirectX documentation)and force anisotropic filtering then we do trilinear anisotropy on the first stage and bilinear on the rest.

It is not easy to turn those optimizations off

On the contrary it requires an act of will and some additional effort to turn them on.

... there was a performance hit on the R420.
Most reviewers overlocked that fact. They just benchmarked and did not analyse this matter.

I suggest that if you want to make statements like this then you actually go and collect information from the reviewers and enquire how they enabled anisotropy in each case in order to confirm your speculation so that we can see if you do, in fact, have a point.

Note that in ATI have repeatedly stated that the correct way to enable anisotropic filtering is always using the application's own settings, where provided, and that the control panel should only ever be used for legacy applications where no such options are provided.

 

[WaltC]

So, are you accusing Dave of spreading misinformation, or suggesting Voodoo2 doesn't support multi-texturing?

Nope... I'm accusing anyone who thinks the Catalysts *always* force bilinear on the first stage when multitexturing of either never having read what Dave wrote, or else of never having comprehended it... Pretty simple--when the Cat CP is set to "Application Preferences" the application makes those decisions, and if the application says tri on the first stage--or any other stages or combination of stages--when multitexturing, then tri it is. That's what Dave wrote, btw.

This is wholly unlike the Forcenator setup in which bri is forced whether the CP is set to "Application" or whether it isn't, and from what I read of the newer Forcenators, bri is the standard default for Application and CP control alike, and you have to turn it off via an option switch in the CP, which is an improvement over the earlier Forcenators, certainly. Problem is, though, that it isn't clear that the "bri defeat switch" actually always works in every case when used, from what I've read of it so far.

The V2 is entirely your concern.

 

[WaltC]

Is there really any purpose to this thread anymore? Chalnoth hasn't even replied to the masses of responses. . .

Well, it's kind of hard to talk when your foot is in your mouth, I'd imagine...

 

[Richthofen]

bottom line is this:
Note that in ATI have repeatedly stated that the correct way to enable anisotropic filtering is always using the application's own settings, where provided, and that the control panel should only ever be used for legacy applications where no such options are provided.

ah ok i see. Then i will repeatedly state that as long as an application does not request 4x 6x 8x AA or higher ... hm... 2x or no aa is enough right?
Your words.

Bottom line remains. They are cheating and nothing else. Glad that at least one website turned their optimization crap off and benchmarked with fair and comparable settings.
Did not read the beyond3d review that's why i can't say how it was done here but the rest simply benchmarked not under the same conditions.

 

[martrox]

 

[Quitch]

I hate everything that dares enter my systray & run in the background with a passion...

that doesn't exactly warrant another systray icon, another process, another few megs... :?

Radlinker does non of those things

Radlinker all the way!

Bottom line remains. They are cheating and nothing else

How are they cheating? You choose a High Quality option in the control panel not a Trilinear option, they never claim to do Trilinear on all texture stages, but when the app requests Trilinear on all stages, it gets Trilinear on all stages.

It's about as far from cheating as can be.

 

[Evildeus]

 

[Bjorn]

How are they cheating? You choose a High Quality option in the control panel not a Trilinear option, they never claim to do Trilinear on all texture stages, but when the app requests Trilinear on all stages, it gets Trilinear on all stages.

It's about as far from cheating as can be.

I agree that it's not cheating but i still don't understand why they do it. Ok, to gain speed of course but was/is there a problem with adding a checkbox for enabling trilinear on all stages ?

And if it's not requested by the app (as in forced by the CP), does the NV4X f.e do trilinear on all stages while the R420 don't ?

 

[Razor04]

bottom line is this:
Note that in ATI have repeatedly stated that the correct way to enable anisotropic filtering is always using the application's own settings, where provided, and that the control panel should only ever be used for legacy applications where no such options are provided.

ah ok i see. Then i will repeatedly state that as long as an application does not request 4x 6x 8x AA or higher ... hm... 2x or no aa is enough right?
Your words.

Bottom line remains. They are cheating and nothing else. Glad that at least one website turned their optimization crap off and benchmarked with fair and comparable settings.
Did not read the beyond3d review that's why i can't say how it was done here but the rest simply benchmarked not under the same conditions.

You do realize who you are talking to here? Right?

 

[Bjorn]

You do realize who you are talking to here? Right?

It would be a sad thing if people on this boards just agreed on everything one guy said just because he happened to work for one of the IHV's. (perhaps i need to repeat that i don't think that Ati is cheating).