Egg on Nvidia's face: 6800U caught cheating?

[DemoCoder]

The Max Payne image is wierd. Contrary to the explaination given (changed mip level), we have a few differences:

1) all three images are rendered with different frame geometry (look at Max model on RefRast vs ATI for example)

2) if that weird green square were simply a different mip-map being selected, then why does it seem to be blended over background miplevels, why does it have this shape, and why isn't this square visible in other photos (just with a different mip level)

Clearly, this is a quad of geometry being rendered which simply isn't in the other frames, or in a different place

And if the drivers are cheating, why only this quad? The mip levels for the entire rest of the image seem about the same. I'd expect big differents in the whole scene.

No, something much stranger is going on.

 

[FUDie]

But isnt the cliping plane what defines what will be rendered and what wont?
in that case geometry would be dissapearing from the scene if you reduce them, thus have a significant impact on the image? hehe what with the missing stuff and all...
I didnt follow the latest driver cheating very closly since Im not using NV hardware atm..

You use the clip plane to clip out something that you know won't be visible. Then disable (or move) the clip plane so as not to remove things that will be visible. This doesn't work in the general case because you won't know what will be visible and what won't be: That's why we have Z buffers.

-FUDie

 

[FUDie]

The Max Payne image is wierd. Contrary to the explaination given (changed mip level), we have a few differences:

1) all three images are rendered with different frame geometry (look at Max model on RefRast vs ATI for example)

Same location, different animation frame. BFD.

2) if that weird green square were simply a different mip-map being selected, then why does it seem to be blended over background miplevels, why does it have this shape, and why isn't this square visible in other photos (just with a different mip level)

You don't have much imagination, do you? Looks like it was an alpha blended texture to me and that the first mipmap is green.

Clearly, this is a quad of geometry being rendered which simply isn't in the other frames, or in a different place

Nope, it was there, just transparent.

No, something much stranger is going on.

Only because it's NVIDIA.

-FUDie

 

[dksuiko]

no, occlusion culling is not cheating.
why do i feel like i am talking to a brick wall here?
Occlusion culling is IMPOSSIBLE to see a difference in the rendered image, because THERE IS NO DIFFERENCE IN FINAL OUTPUT. Ergo, it is 100% IRRELEVANT to the discussion at hand.

So then is ATI's AF implementation cheating? What about nVidia's Quincunx? These are simply trade-offs for performance with a sacrifice in image quality. I don't see why you were so quick to jump jolle's back and twist his words with that aggressive little reply of yours - he had a point. And I think it's valid: When you have to dig really deep and search hard for the degradation in IQ, then it's probably not worth the bitching. Now, before you go off and say that I'm sticking my head in the ground like an ostrich, I'm not saying to completely ignore obvious cheating like nVidia previously did when 3DMark '03 came out. In fact, this point is brought up in the article in question. With regards to Max Payne 2, the author wrote:

"When your playing away at 100+ fps I have to be honest and say that its not a noticeable change in IQ over the Radeon. You’d be hard pushed to say which image is optimised with the mipmap square. To me though, its not a matter of what I can see so much as the fact that this changing of textures is happening behind the users back." There are acceptable trade-offs and unacceptable ones - it all depends where you draw the line. Just because jolle draws that line in a different spot when it comes to 'optimizations' than you do, doesn't mean he's sticking his head in the ground.

 

[DemoCoder]

Same location, different animation frame. BFD.

Shouldn't people who make a big deal about comparing screens to "how the refrast does it" care that the submitted geometry and pipeline state requested was identical.

You don't have much imagination, do you? Looks like it was an alpha blended texture to me and that the first mipmap is green.

Then why don't I see this texture in the other frames?

Nope, it was there, just transparent.

100% transparent, eh? So one mip-level is 100% transparent, next isn't?

Boy, this is an uber-cheat! I bet this must be responsible for the huge performance gains of the NV40, not the 4x pipelines, 2x bandwidth, 2x shaders, angle dependent AF, and bigger register fifos.

Nope, it's all down to selected use of changing mip levels on single transparent quad. Couldn't be a bug or other artifact, nah. Nvidia's really tricky. They made every other triangle in the entire frame have nearly identical mips to the refrast/ATI, but we caught them and their little green quad!

 

[jvd]

IF you want to offer more modes fine do it. Put them in the control panel. But state exactly what your drivers are doing to us.

 

[sonix666]

Oh my God, they have a difference in LOD compared to ATi. It is shifted a whole 20 pixels!!! How dare they! cheat! Cheat! CHEAT!

Now seriously. This whole thing is again trying to bitch over nothing. The image quality difference aint visible without your colored mipmaps, and the difference in performance ain't gonna be 50%. So far, looking at REAL screenshots, there seems to be no problems with GF6 series, except for visible quality problems in Far Cry.

 

[Evildeus]

"Waiting for FM response" 8)

And as someone statated on another forum, a noob who colors mipmaps and uses photoshop is still a noob

 

[Althornin]

So then is ATI's AF implementation cheating? What about nVidia's Quincunx? These are simply trade-offs for performance with a sacrifice in image quality. I don't see why you were so quick to jump jolle's back and twist his words with that aggressive little reply of yours - he had a point. And I think it's valid: When you have to dig really deep and search hard for the degradation in IQ, then it's probably not worth the bitching. Now, before you go off and say that I'm sticking my head in the ground like an ostrich, I'm not saying to completely ignore obvious cheating like nVidia previously did when 3DMark '03 came out. In fact, this point is brought up in the article in question. With regards to Max Payne 2, the author wrote:

"When your playing away at 100+ fps I have to be honest and say that its not a noticeable change in IQ over the Radeon. You’d be hard pushed to say which image is optimised with the mipmap square. To me though, its not a matter of what I can see so much as the fact that this changing of textures is happening behind the users back." There are acceptable trade-offs and unacceptable ones - it all depends where you draw the line. Just because jolle draws that line in a different spot when it comes to 'optimizations' than you do, doesn't mean he's sticking his head in the ground.

No, because ATI's AF and Quincunx are options. hello? anyone there? "Aggressive little reply" - nice belittling little comment - And i didn't TWIST anyone's words, at all.
as per your second paragraph, i refer you to this (part of my post you aparently didnt read):

I feel that what is "noticeable" differes from person to person, and I want to be able to make the call on noticing it or not - not have it made for me.

So whats your point? You agree with me?
As for sticking his head in the ground - thats because of another thing i already mentioned (but far be it from you to read all of my post, eh?)

I also feel that giving a carte blanche on "unseeable optimizations" is a quick step down a veeeery slippery slope, and i dont want to go down it.

I think ignoring that potential problem i*is* sticking your head in the sand.

 

[Bjorn]

No, because ATI's AF and Quincunx are options. hello? anyone there?

So ATI's "angle dependant" AF is an option ? (still remains to bee seen if it's optional on the NV4X)

 

[Althornin]

No, because ATI's AF and Quincunx are options. hello? anyone there?

So ATI's "angle dependant" AF is an option ? (still remains to bee seen if it's optional on the NV4X)

yeah - AF is an option. You know, you can turn it on and off?
And the effect that said toggle does are not in any way hidden from the consumer. SO yeah, its an option

 

[Bjorn]

yeah - AF is an option. You know, you can turn it on and off?
And the effect that said toggle does are not in any way hidden from the consumer. SO yeah, its an option

It's pretty obvious that it's optional to use. But that doesn't mean that we shouldn't care how it's implemented.

 

[Evildeus]

But is i want to use AF, the depend part is not an option is it? (not that it really matters for me )

 

[Althornin]

yeah - AF is an option. You know, you can turn it on and off?
And the effect that said toggle does are not in any way hidden from the consumer. SO yeah, its an option

It's pretty obvious that it's optional to use. But that doesn't mean that we shouldn't care how it's implemented.

i do care - as should be obvious if you read these boards - but you are missing the point. Turning on AF/AA is an option. You know exactly what you are getting - be it quincunx or ATI's "angle dependent" AF. Its not even the same thing. you people are trying to do something (i guess convinnce me that if you cant really see it, its ok to let is slide??) by using really poor and totally flawed analogies.

Changing LOD with no option to fix it and without letting the customer know is completly different from offering an option that has advantages (and not so big advantages - ie, the less AF'ed angles). I dont see how you can fail to comprehend this, which is why i am getting snippy.

Any non-option addition that degrades IQ is not acceptable, especially if it does it only in certain games (IE, if the card cannot do real trilinear, thats fine - consumer isnt being tricked - but if it can and doesnt, thats not ok - ie, brilinear on 5x00) while an option whose behavior is totally well defined is a consumer choice (ie, AF, be it ATI's or nVidias, and with the 6800, brilinear/trilinear)

 

[PatrickL]

What they are doing is just trying to derail the thread on purpose, as often lately.

 

[dksuiko]

No, because ATI's AF and Quincunx are options. hello? anyone there? "Aggressive little reply" - nice belittling little comment - And i didn't TWIST anyone's words, at all.
as per your second paragraph, i refer you to this (part of my post you aparently didnt read):

I feel that what is "noticeable" differes from person to person, and I want to be able to make the call on noticing it or not - not have it made for me.

So whats your point? You agree with me?
As for sticking his head in the ground - thats because of another thing i already mentioned (but far be it from you to read all of my post, eh?)

I also feel that giving a carte blanche on "unseeable optimizations" is a quick step down a veeeery slippery slope, and i dont want to go down it.

I think ignoring that potential problem i*is* sticking your head in the sand.

Sigh.. see what I mean? Do you have to put that shit "Hello? Anyone there?" in your posts or can you just not resist the temptation to be an asshole? Jeez.. and labeling your post 'aggressive' is belittling? In any case, when you inject your posts with such a trolling attitude it makes for an unproductive discussion.

A discussion I'd rather not participate in. No fun to me. Thanks. Bye.

 

[FUDie]

Same location, different animation frame. BFD.

Shouldn't people who make a big deal about comparing screens to "how the refrast does it" care that the submitted geometry and pipeline state requested was identical.

I think you are being overly picky because this shows NVIDIA in a bad light. Obviously it's the game rendering a slightly different animation frame. Why not focus on the rest of the scene that is static, hmm?

You don't have much imagination, do you? Looks like it was an alpha blended texture to me and that the first mipmap is green.

Then why don't I see this texture in the other frames?

Because the mipmaps are colored, the base textures are unchanged.

Nope, it was there, just transparent.

100% transparent, eh? So one mip-level is 100% transparent, next isn't?

I have no idea how the colored mipmaps were generated, maybe they replaced alpha too, hmm?

Boy, this is an uber-cheat! I bet this must be responsible for the huge performance gains of the NV40, not the 4x pipelines, 2x bandwidth, 2x shaders, angle dependent AF, and bigger register fifos.

What does this have to do with the discussion? Stop trying to distract people from the issue raised by the images.

Nope, it's all down to selected use of changing mip levels on single transparent quad. Couldn't be a bug or other artifact, nah. Nvidia's really tricky. They made every other triangle in the entire frame have nearly identical mips to the refrast/ATI, but we caught them and their little green quad!

Sure it's a bug. Yet the same LOD bug is shown in all screenshots. Maybe there's some weird bug with the drivers causing that one quad to be not transparent, but that doesn't explain the LOD shift elsewhere.

-FUDie

 

[ChrisRay]

I'm curious why this hasnt been asked, But has anyone with the 6800 U actually tried toying with LoD to see if it can be changed? Even Riva Tuner can manipulate LoD. To see if its a bug ect ect.

Guess it'd have to be asked to one of the few reviewers with the card, Thats the problem with info Like this. Its very hard to confirm or test, When the card hasnt even been released yet.

 

[DemoCoder]

Sure it's a bug. Yet the same LOD bug is shown in all screenshots. Maybe there's some weird bug with the drivers causing that one quad to be not transparent, but that doesn't explain the LOD shift elsewhere.
-FUDie

There is no LOD shift elsewhere in that Max Payne shot. Only the quad is significantly different. Should I say BFD like you said to the issue of differing frames? You seem perfectly willing to dismiss the different pipeline state and geometry as BFD, but when every there LOD level is basically identical except for a single anamalous quad, we're supposed to see it as evidence of cheating.

 

[Ardrid]

Wtf is with this elitist asshole nature some ppl have on this board? Give the shit a rest. It's no wonder new ppl don't stay long. Is it so hard to be respectful to others in your posts?

As to the issue at hand, I'll wait until more research is done before I make any definitive conclusions.