Has Nvidia Found a Way to get a Free Pass in AA comparrisons

[Hellbinder]

This specific review was covered in another thread. Its the Velocity3d review i am going to reference. What i am wanting to look at in detail is this statement.

Just to make life a little more interesting you can pretty much ignore these anti-aliasing comparison shots! Why? Well because it seems the GeForce FX has added some kind of post frame buffer processing to its list of gifts. By that I mean that in usual cases you can grab a screenshot with your anti-aliasing enabled and the image that gets snatched straight from the frame buffer is an accurate reflection of what you're seeing on your monitor. Not so with the FX. With the FX it seems that 2xAA and 4xAA are both processed after the scene to be rendered has left the frame buffer en-route to your screen and the captures image has no anti-aliasing evident at all.

It would seem that Nvidia, in one released Statement to the various websites, have found a way to basically circumvent the IQ comparrison process. By Claiming that there is some "Secret" "proprietary" unverifiable post process being done on the 2 most common and heavily benchmarked AA modes.

The Question. Is this Legitament??

Anyone with eyes can see that the 2x and 4x modes are clearly behind the power curve when directly compared against the R300's AA modes. Yet Nvidia is claiming a post process Filter effect that "improves" the IQ for the final product that cannot be seen in screen shots. Quincunx for years has used a post filter Blur. Yet it is visible in screen shots. How is it that Nvidia has a new mysterie method of AA filtering that is supposed to Enhance the Quality of the AA? What could they be doing in the final few steps before the Frame hits the screen that will significantly, or noticably improve the AA.

It seems pretty shoddy to me at best. Especially When I am already reading posts at Nvnews and [H] that post the above Quote to say that the Final AA of the Nv30 is actually better than the 9700pro. Without one single thread of verifiable evidence. "you cant consider the AA results" "They dont count" etc etc etc...

How is one supposed to Verify the results of this? What could Nvidia possibly be doing other than Blurring the screen up a bit before final display? How does this constitue "better" looking AA? It seems that Nvidia is going out of its way to not only change the meanings of estabished terminology, but also Change the rules on Fair comparrisons between the competing cards.

 

[maskrider]

That means it is like what 3dfx has done with the V5, the final AA blending is done with the RAMDAC and a special handling for AA modes in HyperSnap-DX are required to take the screenshots.

Is that really the case with NVIDIA ?

 

[tb]

Run a fsaa test in a windowed mode(RAMDAC or other post filters won't work or have to be emulated by the driver) and grap the front buffer and save it to a file.

The gf4 had the same problems, a simple "screen grab" wouldn't work...

Thomas

 

[RussSchultz]

3dfx had the same thing with the V5.

There are two physical buffers that contain the two sets of samples. The RAMDAC grabs samples from each and combines them before outputting to the monitor.

This avoids having the GPU do an explicit downfilter operation and saves that bandwidth.

Why they don't have some method of providing screenshots, I don't know, but the method of FSAA is valid and if capturing a screenshot only returns one of the buffers then it will not be an accurate representation of whats on the screen.

 

[Brent]

HyperSnap DX can capture the screenshots correctly

 

[Hellbinder]

So does the end result improve the Quality? or Blurr the edges more...

It it something that everyone should be doing? or what...

 

[kyleb]

according to the hardocp preview a while back, at least if i rember right; nvidia contaected them about the isisue but claimed it was only with the x2 and said they would find a way to resolve the problem. shortly there after the screenshots were repleced whith ones confermed to be indictive of the actual immage quality. it seemed like an open and shut case to me, i can't help but think that it is just uneducated people that insist *n looks just as good as *n regarless of the hardware in question.

 

[BoddoZerg]

I do sometimes find it hard to believe how awful the 6x and 8x modes look. If only someone who actually had a GFFX could comment on whether 8x nVidia FSAA looks better or worse than 4x ATi FSAA.

 

[andypski]

The problems with making statements like this seem to be numerous -

- You can claim whatever you want.
- There is no way to verify the claims.
- How can anyone know if HypersnapDX (or any other app for that matter) is actually truthfully representing the results of the post-filter.
- What should a journalist do if the manufacturer doesn't provide them with some way to accurately grab the frames? Should they even mention the mysterious post-processing if they have no evidence of its facility?
- If a given hardware manufacturer X gives you some software or a library to 'accurately' capture the screen image, how can you be sure that the software is doing exactly the same thing as the hardware? There is time for much more advanced image processing in an off-line app than in real-time hardware.
- If they are completely on the level then how can they prove this to you without revealing details of the protected 'proprietary' technology that they would rather keep secret. If the technology is really that good then there will no doubt be all sorts of agreements to sign to get access to the screen capture technology, including ones that prohibit reverse engineering to find out what is really happening.
- If they're not on the level then how do you go about exposing this? Again you are likely to be prohibited from reverse engineering the screen capture program.

To quote -

"Can open, worms everywhere..."

Of course, you could choose to just trust what you're told... surely nobody would deliberately mislead people?

 

[RussSchultz]

I really can't come up with a way to say it without being slightly inflammatory, so you'll just have to deal with being offended.

Are you sure you're not in the marketting department? That was a perfect example of FUD. Great use of facts that insinuate that there is something underhanded going on, plus the little nudge at the end to push us over the edge. Kudos.

Seriously, as a competitor's employee, you shouldn't make comments like that. After I finished reading it, I felt dirty.

 

[Joe DeFuria]

Seriously, as a competitor's employee, you shouldn't make comments like that. After I finished reading it, I felt dirty.

Rsading official and misleading product spec sheets, on the other hand....no problem.

 

[kyleb]

odd RussSchultz, here i though that andypski's comments were compleatly level headed and of great importance to the topic at hand, regardless of what he does for a liveing. as Joe DeFuria eluded to, at the very worst it would be a case of the pot calling the kettle black.

 

[Althornin]

Rsading official and misleading product spec sheets, on the other hand....no problem.

Of course not!
the final performance is all that matters

 

[RussSchultz]

Try to keep your comments in the thread from which they originate.

Plus, at least try to quote me in context.

 

[OpenGL guy]

Next we'll see SIS claim that screenshots don't accurately reflect what their "turbo texture" modes and their "jitter-free jittered AA" do.

 

[kyleb]

Try to keep your comments in the thread from which they originate.

Plus, at least try to quote me in context.

i lost you there man, could you explain that a bit beter?

 

[RussSchultz]

Sorry, it wasn't directed at you, it was directed to Joe and Althornin.

 

[Joe DeFuria]

Russ,

All I'm looking for is a bit of consistency here.

1) Official spec sheets on product boxes that are misleading at best and may not tell the "whole story" = Who cares, no need to rant.

2) ATI employee not speaking for ATI makes points that are valid concerning the ability to truthfully accurately represent the product = you rant.

 

[andypski]

I really can't come up with a way to say it without being slightly inflammatory, so you'll just have to deal with being offended.

Are you sure you're not in the marketting department? That was a perfect example of FUD. Great use of facts that insinuate that there is something underhanded going on, plus the little nudge at the end to push us over the edge. Kudos.

Seriously, as a competitor's employee, you shouldn't make comments like that. After I finished reading it, I felt dirty.

This is going to be quite long I'm afraid...

Anyway, it's ok, Russ - I'm not offended. I don't mind you pointing out these things at all, and as a known competitor's employee on this board I do try to be very careful about what I say, and usually talk in general terms.

If this post seemed a bit too 'marketing' then I apologise - I guess I may have been emotionally influenced by other things that I have been reading here and elsewhere. I shall try to clean things up in this post.

To clarify for yourself and others who may be concerned - I personally believe that HypersnapDX is currently capturing representative information. I also believe that it was important and correct for nVidia to point out mistakes made in the reviews where the images were not being captured correctly.

In a perfect world it should have been immediately clear to people that the 2x and Quincunx AA screenshots were not properly representative of their output since no AA was apparent in the shots at all despite the altered performance. The operation of the Quincunx filter from previous hardware is pretty well understood - I don't know is if GeforceFX's post-filter is the same or modified, and if so whether HypersnapDX has been modified to represent whatever enhancements took place, but if nVidia are stating that the output is now correct then I expect that to be the case.

To address this further - what real guarantee do you have of any screen capture - no matter what the method? At some point, to get at the data you have to go through the driver (ours or a competitors), and it could be invisibly manipulating the data that you get back through pretty much any path.

If I did spread some FUD in my post I don't think that it is any that shouldn't be in people's minds anyway when they see vague statements about unverifyable quality improvements. It is one thing to give details on an algorithm, but quite another to say 'it just looks better'.

Verifiability of some sort is key, and needs to be provided, but in a secretive industry such as ours it can be difficult. If an algorithm can be clearly described then people can debate its merits openly. If an algorithm is secret there can be no debate - you either believe the claims or you don't. Unfortunately revealing algorithms can put you at a competitive disadvantage since it allows people to catch up with what you are doing - for example we could have either said :

"We have secret proprietary technology that makes our AA look better, sample for sample, to our competitors."

... or we could say :

"In AA modes we apply sub-sample gamma correction to give a more appropriate response curve to sample values and improve their quality"

I do know that we have tried to be open and give pretty full descriptions of the techniques used for our antialiasing, including details such as the subsample gamma correction and programmable grid.

If nVidia have had problems it was because they didn't provide an appropriate method initially to get the correct screenshots, and hadn't briefed people to expect these problems and the cause behind them on day one. Some level of uncertainty since then about their methods should be expected, and they aren't the first people to suffer rightly or wrongly from this.

I hope you feel better about this post.

- Andy.

 

[RussSchultz]

I appreciate your followup. Thanks Andy.

I made the statement, even though it would draw flames from some folks like flies on stink, because I'd prefer to see all of the IHV posters here remain apart from the partisan accusations and silliness. I particularly don't want to see this board becoming a staging area for different IHV marketting groups to wage their wars.

While I agree, there is a potential for subterfuge and we essentially have nothing to rely on but to trust NVIDIA that their snapshots are doing the right thing, your post just struck me as a unfortunate diversion from the usual modus operandi.

Plus, I'd hate to see it blow up into something bigger and risk your presence here (along with your co-workers). Imagine if the Inquirer picked up on it and ran a story "ATI employee Andy says NVIDIA's plugin is full of poop".