Why do we still have fixed function AA and filtering?

Mixing 4xOGMS with 4xOGSS in order just to claim 16x sample AA is senseless IMHO. You get almost equivalent output from 8xS and better performance.

Even with 16xOGMS you fall behind in terms of edge quality compared to a 6x sparse sampled grid.

 

I don't believe sparse sampling does quite that much for you. This is definitely falling into the region where the angles at which 6x sparse looks poor should be more noticeable than those where 16x ordered looks poor.

 

That's very true, thanks a lot for remembering me of that! It's so simple I'm ashamed I didn't realize it myself...

 

Like I said...

I would put $$ on the table that a very small number of people actually use those modes...

Personally, I would find it somewhat compelling if the ends justified the means, so to speak. In other words, if ATI offered some 8/16 sample mode that chugged along...but looked incredible...I would consider the tradeoff for certain games (especially flight sims). But I have a very hard time seeing the visual benefit of nVidia's modes.

 

I did use 8xS (1024*768) in FS2002 f.e. on the NV25, so please don't shoot me if I'm an exeption here. The only other reason was that it's an abysmally CPU bound flight sim and the 4*4 grid was the best available choice.

Don't ask about impressions today

 

3dlabs latest vpu supposedly supports programable AA and Filter I dunno how it is exposed if it is.

 

Yeah I missed to mention little important details like bandwidth and resolution. Inevitably a 6 sparse grid MSAA allows you much higher resolutions than 16xOGMS will (all concentrated on the Z units/bandwidth current high end hardware has).

(1024*4) * (768*4)
vs
(1280*6) * (960*6)

Just a simple example.

Besides I still doubt that the claimed 16x mode is anything else but a hybrid sampling mode, ie bye bye to 1/4th of the fillrate.

 

Yes, that does make sense. But the resolution difference would be much, much more significant.

Anyway, I would like to say that when somebody says, "when looking at so and so," I tend to just consider that effect, not any secondary effects (in this case, texture AA and performance).

 

I'm not so sure what you really mean with "secondary effects" especially when it comes to performance.

Performance has all the relevance in the world when it comes to comparing different kind of implementations, especially if there's supersampling involved.

Paradigm A:



Paradigm B:



It is that irrelevant in reality

 

just a small question.

Why do the guns in nVidia screenshots always look better than the ATI screenshots?

Totally irrelevant to the perfromance trade off comparison here, but its been noticed a lot. Its not down to LOD bias is it?

 

All effects are relevant in the end result.

But you just said "edge quality," not performance, not texture quality, etc.

 

I think in this case it's probably due to the use of 8xS, which makes use of supersampling, and therefore will increase texture clarity across the entire scene.

But, I do have to say that upon close inspection, the lighting looks very slightly different on the gun in the two pics. The gun in the nVidia pic looks like it has more red light. I don't know if this is due to slight differences in position, or of it's part of what makes it look better. But that difference is worth noting.

 

That's easy:

(A) gamma callibrated
(B) default

Niet

The 2nd shot has 16xAF.

 

A. 16-degree anisotropic filtering will only have an effect on surfaces that are angled at greater than an 8:1 ratio in some direction. The vast majority of surfaces are not. Supersampling affects all surfaces, allowing for more aggressive LOD without aliasing.

B. 4x supersampling (as is used with 8xS) achieves essentially the same effect, for high-angle surfaces, as doubling the degree of anisotropy. Thus, for surfaces that would use 16-degree anisotropy on the Radeon, the GeForce FX's 8-degree anisotropy + 4x supersampling achieves similar results.

Anyway, look closely at the rough part at the top of the gun. I get the impression that the FX picture is very slightly more clear, which would be in line with what is expected from supersampling.

Final note: I don't feel supersampling is a good use of computational resources. The effect on textures really isn't huge, as there are better ways of improving texture quality (better=higher visual quality/performance ratio). Still, it may be nice if sometime in the future, it becomes feasible to have, say, 4x SSAA always on (OGSS would be fine, as long as it's combined with a more advanced MSAA technique for edges...OGSS is quite possibly better than more erratic sampling techniques for textures). At the same time, I think I'd rather just have bicubic sampling for magnified textures first (should be similar in math requirements to 4x supersampling, but only required for a part of the screen).

 

Is it just me or is the nVidia shot *more* aliased on certain angles than the ATi one?

 

Aliasing is notoriously difficult to pick out in static screenshots. You typically need regular patterns to pick out aliasing. So, in essence, you may be right, but I wouldn't bet on it.

 

I would bet on it.
Static screenshots like the pinwheel type that we have seen in several reviews show that ATI's 6x has better edge quality than nVidias 8x

 

If you zoom in, in certain spots yes and in certain other spots no. Here's to note that it's 4x sparse on the Radeon and not 6x as some might believe. I picked on purpose scenarios that don't exceed a 4*4 grid.

Mipmapping methods aren't even completely identical between the two cards; 4x SSAA affects surfaces in the same degree only up to a specific part of the scene, in any case not further than Level 2x Aniso will. In realtime you get extensive blur in the scenery in the far clipping distance. Usually IHVs offset LOD for 4xSSAA to a usual -0.5 degree. Big deal if you ask me.

Not sure what to make of that one.

A= 8xS/ no aniso
B= 4x MSAA/16x AF

Adding Level8x to 8xS would be an effective fillrate/bandwidth suicide.

You can sit there and overanalyze the two shots to oblivion. I had to pick 800x600, since otherwise with 8xS in a higher resolution I would have been dumped back to the desktop, especially in that map.

The game isn't even playable with 8xS on, hence....

Now we're in agreement. It was merely an illustration concerning edge quality and the connected performance and feasability of each method. In retrospect both cards used have about the same effective multitexturing fillrate and it's always better to be able to crank up the resolution with edge/texture anti-aliasing combined, than being bound to resolutions that remind the V2 era.

 

Sorry, for some reason, I was thinking of texture aliasing.

Anyway, 8xS vs. ATI's 6x should be relatively similar in edge AA quality, as far as the sample patterns are concerned (each one should win at different angles). One crucial difference, however, is that ATI's uses a gamma-correction algorithm, which will skew it slightly in ATI's favor.

 

Sorry, didn't notice that. Yes, on any surface at higher than 2-degree, the ATI card should be showing better texture clarity (except for those angles which the ATI card's aniso algorithm is poor...). I don't know what degree that part of the gun I pointed out is, but chances are it's under 2-degree...