[Rancidm]

http://www.digit-life.com/articles/parhelia/index.html

Pretty indepth parhelia review. The only thing that bugged me was they used the 8500, geforce4, and parhelia for the entire article.. but they left out the 8500 in the ansiotropic filtering tests.

[Dolemite]

Yeah, and they still insist that the 8500 is doing RIP mapping, which has been clearly demonstrated to not be the case.

In any case, the mere existence of different methods of anisotropic filtering only validates the demonstration of those methods even more.

IMO, they're playing favorites here by not testing either the performance or image quality of the 8500's aniso.

[Dolemite]

From the review:

Have you noticed that below on the left the border of the pistol's handle is not processed in FAA 16x? Such artifacts are typical not only of this test. Well, it reminds me of the ATI's anisotropy.

Ooooh, well maybe you shouldn't have tested FAA since it uses a different method as well. nVidia is the standard by which all others are judged after all..., only their AA and aniso methods are valid...others are just cheap hacks.

[Bambers]

Maybe someone should show them some alpha tested textures. (theres quite a few in JK2).

[Dolemite]

Another thing I noticed is some strange artifacting in the closeup GF4 FSAA shots:



It looks like they could just be compression artifacts, until you realize that no such artifacts are seen in the Parhelia shots, which you would expect to see if both shots were compressed at some point:



And yet the Digit-Life reviewer says nothing of this glaring difference, but goes on to say how the Parhelia produces artifacts by missing certain edges. WTF? If he's not an nVidiot, he's certainly a piss-poor reviewer/technical journalist.

[Basic]

I agree that the reviewer is sloppy to let through compression errors like that. Images should be saved in lossless format, be magnified, and then possibly be compressed. Never magnify a destroyed image when you have the original.

But OTOH, it's quite often that happens. So it certainly isn't anything special for this reviewer. And it's kind of funny that you call him a nVidiot when he shows images that makes a GF4 look bad in a unfair way.

A small jpeg compression error lesson: Download

[muted]

The performance of vertex shaders with the promised 4 pipelines is too low. I think that only one of them really works. It's possible that lack of the DirectX 9.0, and therefore, the version 2.0 could cause locking of 3-4 pipelines. Next time we will check this hypothesis

could this be possible ?

I guess we'll just have to wait til someone does some test using DirectX 9.0 ( don't know who would venture and install the beta flying around )

[ben6]

The conclusion is incorrect on the vertex shaders

[Dolemite]

Originally Posted by Basic

I agree that the reviewer is sloppy to let through compression errors like that. Images should be saved in lossless format, be magnified, and then possibly be compressed. Never magnify a destroyed image when you have the original.

But OTOH, it's quite often that happens. So it certainly isn't anything special for this reviewer. And it's kind of funny that you call him a nVidiot when he shows images that makes a GF4 look bad in a unfair way.

A small jpeg compression error lesson: Download

Heh, well I call him an nVidiot because the GF4 images are so clearly inferior and yet he says nothing of it. I do suspect that its just magnified JPEG compression rather than actual output of the GF4, but still...

[Gollum]

Concerning FAA, hop over to 3dcenter.de and check their FAA/FSAA comparison. It's mainly a series of pictures, but interesting to check out! The non-AAed edges with FAA can be VERY irritating to say the least. When looking straight on the edges of polygons, those that get AAed are clearly superior to 4xFSAA, no question, but if you don't look directly at them (e.g. focus on the edge of the picture) and then switch between FAA/FSAA, the difference is night and day, FAA really stands out negatively IMHO. It also made me wonder as to the method behing FAA again, wether the polygon setup in the test can have anything to do with the result?

Of course this is a synthetic test but still, plenty of screenshots in other reviews have similar problems. I'd love to personally see Parhelia in motion to really see just how irritating the introduced pixel popping really is, it might end up not really being noticable in most titles, which I hope. Damn why don't I know anybody who's spending that kind of cash on his GFX card?

Personally I guess I'd rather have slightly worse AA that works on all edges 100% reliably, than better AA that completely ignores some seemingly random edges of the scene, but that's hard to say unless experiencing it first hand ...

[Jerry Cornelius]

Those AA comparisons using FAA don't look like they were done right.

[3dcgi]

[quote="Gollum"I'd love to personally see Parhelia in motion to really see just how irritating the introduced pixel popping really is, it might end up not really being noticable in most titles, which I hope. [/quote]

Why do you think there is pixel popping as oposed to the edge never being antialiased? I agree it would have been nice if an animation of the triangles would have been provided, because motion is the best way to evaluate antialiasing.

[Dolemite]

Some of the other shots definitely don't look compressed, but still show much more graininess from the GF4, even when comparing the same AA method:





To me, this is interesting to note in and of itself and also interesting to note that it goes unmentioned. I usually don't get into these nVidia favoratism arguments, but to me its obvious that this is either sloppy testing or biased reviewing.

[Mintmaster]

I had been waiting for the Digit-life article on the Parhelia for a long time, but this probably has the most incorrect information of any Digit-Life article that I've ever seen.

First, how can they say that about the Parhelia's vertex shaders? Why in hell would Matrox advertise 4 shaders and not get them working in the simplest of tests at launch?

Second, have you ever seen such blatant disrespect for ATI's anisotropic filtering? The few caveats of ATI's implementation don't show up very often, and for the LAST time, it is NOT ripmapping. I want to make a program to clearly show this, but I haven't had the time to pretty it up. I think this article will give me the motivation to finish it ASAP. Leaving it out of the comparison, especially considering how horrible Matrox's quality is, is just wrong.

Third, how can they say drivers will improve the situation for such simple tests? Even the crappiest drivers would not affect these sorts of tests.

There are so many other completely baseless conclusions as well.

Parhelia's seemingly high specs seem only to compensate for poor engineering. Even with the lower clock, its architectural advantages should have made it at least equal to the Ti4600. I suppose it has to do with Matrox having been out of the high performance 3D market for so long, so they didn't have the talent to take proper advantage of so many new technologies. It seems that they have an inefficient memory controller, but since their silicon isn't performing up to snuff for vertex related things, they have serious problems.

On the bright side, R300 is not too far away. By then we'll forget all about Parhelia...

[Gollum]

Originally Posted by 3dcgi

Why do you think there is pixel popping as oposed to the edge never being antialiased? I agree it would have been nice if an animation of the triangles would have been provided, because motion is the best way to evaluate antialiasing.

Maybe I expressed my concerns wrongly (I'm not sure if there is a technical definition for "pixel popping"? ), IMHO when playing without any AA, the jaggies are pretty distracting at anything below 1280x960. When in motion, some nonAAed edges can create some distracting looking aliasing which I meant when refering to pixel popping (in the sense that they just pop into your attention although you're looking at another part of the screen), this depends on game, resolution, models, geometry, texture quality and contrast etc. but a great many games have plenty situations that cry for AA louder than others. If Parhelia would happen to ignore some of those edges (e.g. they're not polygon edges but intersections) during gameplay, then to me personally it'd probably be worse than having an an overall less efficient AA method (that takes care of all edges).

It's really hard to say that without seeing it for real though, I'm trying to be very carefull about this and am well aware that the 3dcenter.de test was synthetic and probably exaggerated the possible differences a lot. Overall I got the impression that FAA works pretty well most of the time and looks marvelous, so until I got to check on my concerns I wont make a judgement on the whole thing, especially since its also a matter of personal preference.

Originally Posted by Jerry Cornelius

Those AA comparisons using FAA don't look like they were done right.

I think they were done just right, not much you can do wrong there. I don't doubt the tester's ability to take the proper screenshots and set up a webpage, nor do I think the test was manipulated to any method's advantage on purpose. Thing is, had they used a different triangle setup for the tool (4 instead of 2 intersecting ones), Parhelia would've probably AAed the intersection, but the purpose of the test was exactly to find out wether such a weakness exists and where it appears, so that's fine. This way we finally know one of the weknesses of FAA (pretty sure anyway) - polygon edges get AAed, but intersections of polygons Parhelia has its problems with. Such edges occur in most games but usually not on any majority of the screen. It occurs mostly on character models I guess, not static level geometry, but that too depends on the game/engine (would go too far IMHO to go into those aspects now)...

[nggalai]

Hi there,

Originally Posted by Gollum

Concerning FAA, hop over to 3dcenter.de and check their FAA/FSAA comparison. It's mainly a series of pictures, but interesting to check out! The non-AAed edges with FAA can be VERY irritating to say the least. When looking straight on the edges of polygons, those that get AAed are clearly superior to 4xFSAA, no question, but if you don't look directly at them (e.g. focus on the edge of the picture) and then switch between FAA/FSAA, the difference is night and day, FAA really stands out negatively IMHO. It also made me wonder as to the method behing FAA again, wether the polygon setup in the test can have anything to do with the result?

Why, thank you for mentioning that article.

Link: http://www.3dcenter.de/artikel/parhelia_aatest/

It's yet another follow-up to the 3DC Parhelia first-look. The main point of interest (and why the article was written to begin with) is to show how the Parhelia deals with AAing polygon intersections. In short, intersections don't seem to be anti-aliased at all, on Parhelia.

ta,
-Sascha.rb