[cellarboy]

From [H]Ocp:

GeForce FX Reviews Wrong?
While we are still looking into this, it seems that the in-game screen shots posted on the Net yesterday showing off IQ produced by the GeForceFX 5800 Ultra are "wrong".

There is no doubt that we criticized the GFFX for its AntiAliasing, and now it seems that we may have not had the proper evidence to base our conclusions on. To quote ourselves from this page:


With NoAA you can see the aliasing is quite predominant. 2X AA and Quincunx don’t seem to do much on the GeForceFX visually, but the FPS are effected comparing the shots to the original with no AA enabled.

Of course all of this left us a bit puzzled, and wondering about the AA abilities of the drivers, but the "facts" are the fact correct?

We have been working with NVIDIA on this to get an answer and it seems that now we have the preliminary information to give us a bit more insight on the question.

The GeForceFX's technology applies filters that effect AntiAliasing and Anisotropic filtering after the frame has left the frame buffer. In short, this means that all of our screenshots do not accurately represent the true in-game visual quality that the GFFX can and will produce, as the screen shot were pulled from the frame buffer. We have come to conclusions about the GFFX IQ (Image Quality) that are simply wrong.

While we cannot answer for other reviews of the GeForceFX it is very possible this is an issue with those articles as well, if they were in fact thorough enough to cover IQ.

We are currently working on a way to capture the images properly and will be revisiting the GeForceFX 5800 Preview by covering the IQ portion of our preview with proper screen shot comparison or further addressing the truth surrounding this situation.

Certainly this is a huge issue it seems that NVIDIA was not even aware of when they issued us the review units. Having 48 hours to preview the card over Superbowl weekend compounded this, and while that is no excuse for improper evaluation on our part, it did certainly impact our ability to do a better evaluation. We are sorry for any incorrect evaluations we have made and are working now to remedy the situation. Any new information will be posted here on our news page.

[Dave Baumann]

http://www.beyond3d.com/forum/viewtopic.php?t=4043

The issues has been evident since GF4 anyway, so why they didn't know about this until now is beyond me. I'd guess that this only applies to 2X and Quncunx, given whats in that thread.

[Joe DeFuria]

Man, the parallels between this and the VSA-100 launch just get more and more creepy!

[Mintmaster]

Soon we'll have to do FSAA comparisons with a magnifying glass and a digital camera...

[OpenGL guy]

From [H]Ocp:

The GeForceFX's technology applies filters that effect AntiAliasing and Anisotropic filtering after the frame has left the frame buffer. In short, this means that all of our screenshots do not accurately represent the true in-game visual quality that the GFFX can and will produce, as the screen shot were pulled from the frame buffer. We have come to conclusions about the GFFX IQ (Image Quality) that are simply wrong.

It's not possible to apply anisotropic filtering to the whole frame at once as you need the texture data, not the frame buffer, to compute the values. I believe HardOCP misunderstood something here.

[Doomtrooper]

The Author of Hypersnap has never been aware there was a need for a Nvidia post filter caption, I've emailed him before on this as I got a copy.
Since Hypersnap is capable of capturing 3DFX's post filter maybe worth a shot to try it..its a very good screen capture utility and there is a trial version.

[Sharkfood]

The issues has been evident since GF4 anyway, so why they didn't know about this until now is beyond me.

Because in the case of the GF4, it allowed them to illustrate how superior Quincunx was compared to the GF3 by using screenshots.

Seriously, I was baffled to see razor-sharp screenshots of the GF4 with Quincunx on all the major websites, yet firing up a game of Everquest or Tribes2- the huds and text were extremely difficult to read and blurry. Sure enough, taking some screenshots myself (and the debates that ensued on this forum) showed the framebuffer snapshots did not truly represent the final on-screen image quality.

How sites could say how much sharper and better Quincunx was based on screenshots only illustrates reviewers need to spend more time on reviews. Had they actually played the games and paid attention to the on-screen, in game IQ, it would have stood out like a sore thumb how different the screenshots are from in-game.

Can't say I blame HardOCP or Anand on this front though. Given TWO (2) days to pack as much information, benchmarks and html as they did is just uncanny. I'd really like to know what caffeine/taurine supplements they use and where I can get some.

[Arun]

Actually, I believe the problem with AA *is* the post filtering.

As for AF... I think nVidia is simply trying to make people think the problem is the same. However, I don't think that's the case. In reality, they probably made their algorithm too aggressive. So, they're going to make it less aggressive in a future revision, increase the performance hit, and no one will notice it because overall performance will have increased with more mature drivers.


Uttar

[nggalai]

I, too, wonder what NV actually meant by stating "The GeForceFX's technology applies filters that effect AntiAliasing and Anisotropic filtering after the frame has left the frame buffer. " Filters? Like, blur filters? Or were they just too lazy to explain the GF4-like "sampling" that happens after the framebuffer?

*puzzled*

ta,
-Sascha.rb

[Bambers]

So are the need for speed screenshots on hardocp ok and the its the ut2003 shots where the AA isnt showing?

[Nazgul]

I'm sure it's possible that the FSAA on the FX is better than the screenshots would indicate, especially if they borrowed some ideas from 3dfx(who also did AA like that as I recall). However, the reviews I read still said that in games the 9700 appeared to do 2xFSAA a good deal better than the FX. I'll be interested to see how this plays out, but I'd place my money on ATI still coming out the winner when it comes to FSAA quality.

[Doomtrooper]

Its still OGMS vs RGMS.

[Joe DeFuria]

Its still OGMS vs RGMS.

Not with 2X. 2X on the GeForce should still be rotated. It could still be of different quality depending on the exact location of samples, and whatver gamma correction (or lack therof) is applied to the FX.

[Chalnoth]

Originally Posted by OpenGL guy

It's not possible to apply anisotropic filtering to the whole frame at once as you need the texture data, not the frame buffer, to compute the values. I believe HardOCP misunderstood something here.

True, but you cannot judge the texture quality without looking at the actual final output. Depending on how the samples are blended, the output blending can increase (supersampling) or decrease (QC/4x9) the texture quality of the output.

[Babel-17]

Lol, so the following statement could be taken as true?

"Screenshots can't begin to describe how good the 2x aa of the FX is."

My feeble attempt at injecting some levity.


I too thought it was fairly well known that screenshots of aa in action couldn't always be taken at face value. Quality in motion is another thing too but that leads us to subjective impressions which is a whole another can of worms.

[Luminescent]

According to Anand ( http://www.anandtech.com/guides/viewfaq.html?i=39 ) 2X MSAA is rotated grid.

[Doomtrooper]

http://www.beyond3d.com/forum/viewto...3&start=20


This thread applies also.

[Lincoln]

Originally Posted by Doomtrooper

Its still OGMS vs RGMS.

The 4xS, 6xS, and 8xS AA modes all have a 2x RGMS component to them, in addition to the OGSS part.

Lincoln

[Doomtrooper]

Yes but we have no problem capturing 4XS prior to this on Nv25, or any of the XS modes...on NV25 hardware.

Here:

http://www.nordichardware.se/artikla...k/2003/nv25AA/

and here:


http://www.beyond3d.com/forum/viewto...54&start=0

[Luminescent]

I believe the capture problem on the FX affects only the 2X mode (correct me if I'm wrong).

[Xmas]

Originally Posted by Lincoln

The 4xS, 6xS, and 8xS AA modes all have a 2x RGMS component to them, in addition to the OGSS part.

Lincoln

I'm not so sure about that. There are two modes of each 6x and 8x in the Detonator drivers. You can use aTuner to enable them.
One 6x mode is 4xOGMS + 1.5x1 OGSS, the other is 2xRGMS + 1.5x2 OGSS.

One 8x mode is 4xOGMS + 2x1 OGSS, the other is 2xRGMS + 2x2 OGSS.

And I'm almost sure that the modes used on GFFX are the ones with 4x OGMS. At least the pics look like it.

[demalion]

The GF FX is not the GF 4. What was true for the GF 4, therefore, is not necessarily true for the GF FX.

Also, my current understanding is that the 2x RGMS component of AA is what is downsampled. Therefore...this would have an effect on 2x RGMS, and the 4xS, 6xS, and 8xS mode screenshots. This seems pretty reasonable.

I presume other possibilities, like Quincunx being incorporated into all modes or some such, would have been noticed when viewed on screen, or are discounted for reasons I'm not thinking of at the moment.

Ack! Xmas, way to throw more confusion into this.

[Lincoln Bauman]

This is my understanding of the "_xS" modes:

2x RGMS + 1x2 OGSS- 4xS
2x RGMS + 1.5x2 OGSS- 6xS
2x RGMS + 2x2 OGSS- 8xS

The last two are discussed (in German) here:

http://www.3dcenter.de/artikel/2002/11-20_b.php

Lincoln

[Bambers]

Originally Posted by Xmas

I'm not so sure about that. There are two modes of each 6x and 8x in the Detonator drivers. You can use aTuner to enable them.
One 6x mode is 4xOGMS + 1.5x1 OGSS, the other is 2xRGMS + 1.5x2 OGSS.

One 8x mode is 4xOGMS + 2x1 OGSS, the other is 2xRGMS + 2x2 OGSS.

And I'm almost sure that the modes used on GFFX are the ones with 4x OGMS. At least the pics look like it.

I'd agree with that. I mentioned it in one of the slew of threads on this (must be something that everyone has to have their own FX thread ). The edge AA and its regularity implies an ordered grid, odd that the choice of the supersampling gives AA emphasis on the verticals though.

[speculation]
The performance hit of the 8x mode in the gf fx also seems very high if its only 2x supersampling? 22% of the no AA speed

A possible reason for these modes being used is that nvidia are having problems with the 2xMS in combination with supersampling (note 4xS and its artefacts) so they had to use the 4xMS in the (p)review drivers
[/speculation]

[Xmas]

Originally Posted by Lincoln Bauman

This is my understanding of the "_xS" modes:

2x RGMS + 1x2 OGSS- 4xS
2x RGMS + 1.5x2 OGSS- 6xS
2x RGMS + 2x2 OGSS- 8xS

The last two are discussed (in German) here:

http://www.3dcenter.de/artikel/2002/11-20_b.php

Lincoln

And the two other 6x/8x modes are described here

http://www.3dcenter.de/artikel/2003/01-17_b.php