Who has the best looking FSAA?

[Pottsey]

In other forums I keep getting told 3DFX have the best looking FSAA with its RGSS method. But no one posts facts.

I dont believe V5 are the best looking one bit as I thought a Radeon 8500 or even Radeon 9700 with Smoothvision 6x Quality with 16x anisotropic filtering and high LOD settings would look far better then a Voodoo 5.

Forgetting about speed, PS, Cubemapping and all those other things missing from a V5. Does anyone know any good articles on FSAA which compare the Radeon 8500 and newer cards with a V5? What about the newer cards like a GF4?

Which card has the best looking FSAA?

The Super-sampling Anti-aliasing Analyzed article link on this site seems to be broken http://www.beyond3d.com/articles/index.php

I know this is a stupid argument as a V5 is good as dead as its far to slow for new games.

 

[Randell]

aaah but with the 8500/9700 and Gf3/4 and Parhelia the AA is augmented by AF. AF is not 'AA' in its accepted sense but does reduce aliasing by better texture filtering. It's also not an even comparison to then say the 8500 must be better with its 6x AA as the V5 5500 only had 4x - the counter is that the V5 6000 had 6xAA and is miles better than 6x SV.

Also are you concerned about purely edge AA or Full screen AA including texture shimmering and pixel popping? (which of course is what AF is used for when multisampling AA is used).

Its also hard to compare directly different methods but with the V5 and 8500 they are both supersampling methods so easier to compare.

At each level of AA (2x, 4x) I can categorically state that the V5 is better than the 8500. I have both at the moment and for shits and giggles I tried UT2003 on my sons V5/Duron box. The quality of just 2xAA was outstanding compared to the 8500 2x or 3x modes.

The Gf3/4's 2x modes are RGMS and offer very good edge AA I hear (better than 4x mode, which is OGMS, in some peoples view). Quincunx offer reasonable edge quality but at a cost of increased blur.

The Gf4 4xS mode is one of the best AA modes available right now (a mix of SS and MS).

Parhelias FAA is also incrediably effective at edge AA at minimum performance overhead and whilst not perfect in its application is getting worked on by Matrox.

the 9700 offers incredible perfromance with 4xAA & 6xAA (which is a jittered MSAA method, read the review here of the 9700) and 16xAF offering excellent IQ.

To sum up, the V5 has definetely been surpassed in overall IQ, but offers the best 2x and 4x FSAA IMO (performance considerations aside) implementation but on hardware with very limited texture filtering capabilities in modern terms.

I'm soory I'm short on facts or links but I'm at work.

 

[Pottsey]

I was thinking along the lines of both cards using 4x FSAA and including texture filtering capabilities like AF.

So which modes on which cards are better then a V5. I take it the Gf4 4xS mode with AF is better looking then a V5?

What about the 9700 and 16xAF is that a lot better then the 8500 with 16xAF?

Thanks for that, its helpful. Now all I need are facts with links. Anyone know of any good articles?

 

[Randell]

Do you know the only person I know for sure has examined Edge AA, texture filtering etc across the V5, Gf2, Gf3/4, 8500 and 9700 with a critical eye is Sharkfood, he may have screenshots, but has not published an overall review.

Typedef has also produced an AA article over at nvnew.net and has a Parhelia.

I dont know if anyone has published anything as all encompassing as you ask for though.

 

[Bambers]

On FSAA alone I would think its very close (well 3dlabs superscene is the best I've heard of by a long way but thats not a games card) the v5 6000 can do 8xFSAA (although very slowly) but the 9700s 6x has its gamma correction and 16xAF for the textures. (the 8500s FSAA is mainly crap, it used to be a RGSS for 2x in one of the first drivers giving good quality but they lost that , oh and the mip levels dont alter properly with AA either).

http://www.beyond3d.com/forum/viewtopic.php?t=2454&start=0 Shows how the gamma correction helps.

 

[Randell]

ooh yes gamma correction - forgot about that. Damn another reason to throw caution to the wind and try out an 9500 or 9700.

forgive me DS

 

[jb]

I was thinking along the lines of both cards using 4x FSAA and including texture filtering capabilities like AF.

So which modes on which cards are better then a V5. I take it the Gf4 4xS mode with AF is better looking then a V5?

No. Two reason. GF4 in x4 drops back to a order grid sampling pattern and looses its ability to deal with edges on horizontal or vertical surfaces. It still does AA on them but it does not have a smooth transistion from AA to non AA. Also Alpha textures while the 4xS mode is a lot better, its still does not AA them as good as a straight RGSS method, ala V5

What about the 9700 and 16xAF is that a lot better then the 8500 with 16xAF?

Yes for the most part, lots better. 8500 could not do tri-linear when it was doing AF and any time you rotated about the Z axis it would stop doing AF at 45 degrees. 9700 seems to over come the 8500 limitations in AF

As far as the AA of the 8500 and V5? I would say 8500 was only with the x6 mode as good or a slight bit better than the V5 x4 mode. However considering that the V5 6000 could do x8 it blows everything else today out of the water consumer wise and only with regaurds to AA...that is if you dont mind effective fillrates at the tnt levels

 

[Colourless]

V56K 8x looks real good, but I can understand why you'd want gamma corrected FSAA like the R9700 does. High contrast lines on a V56K can look kind of jaggy in 8x, when they really shouldn't be. The darkest sample (1/8 brightness) is often 'missing'.

IMO The R9700 6x shots that I have seen look as good as 8x of a V5. Without actually using a R9700 it is really hard for me to say which would really be better, though I lean towards the R9700 at this stage. V5 is just too slow.

Of course there is 2 issues with the R9700 I hear. First one is no FSAA in 16 Bit modes. V5 will obviously kill the R9700 here. Plus generally games that only support 16 Bit are fairly old so the V5 should have enough fillrate for them.

The other is the SuperSampling vs MultiSampling arguement. I really like SuperSampling since is works on Alpha tests, and again this is something that the V5 will beat the R9700 at. Call be spoiled though, as FSAA on means no jaggies anywhere for me.

As always, such comparisons are a subjective thing, and I don't know if comparing with a card that was never released is allowed.

 

[Sage]

well i was an owner of the V5 and i can just say daaaaaaamn. my GF4 Ti still isnt as good as that Voodoo5. oh and the Voodoo5 can do anisotropic filtering... but its slow. Matrox has the best EDGE aa but overalll is say the 9700 would be the best choice

 

[Typedef Enum]

I have recently turned into a Flight sim kind of guy, and can now _really_ appreciate what AA has to offer. Not to say that I couldn't appreciate it before, but flight sims really do expose jaggies in the worst way.

Having said that, I have the following flight sims here:

1. IL2
2. FS 2002
3. F18/A (Janes)

If you really want to compare different implementations, I suggest picking one of these (they are the most popular after all), taking a screenshot (hopefully similar to one another), and then have a discussion.

I could offer up some Matrox fotos later on if anybody else wants to do the same thing (with a different board of course).

 

[jandar]

Sure thing Type...

I have a 9700 and some free time this weekend (wife gone out of town, so time to bum out versus actualy leaving the house)


lets use FRAPS.

I suggest 2X, 4X for both

16x FAA versus 6X AA

8X Ansio


I say MS FS 2002

how about different resolutions as well?

800X600
1024X768
1280X1024
1600X1200

sound good?

I would really like to play with the Parhelia, just cant afford monitors right now.

 

[darkblu]

oh and the Voodoo5 can do anisotropic filtering... but its slow.

any anisotropy seen on vsa100 comes as a result of FSAA supersampling.

 

[Nagorak]

The Gf3/4's 2x modes are RGMS and offer very good edge AA I hear (better than 4x mode, which is OGMS, in some peoples view). Quincunx offer reasonable edge quality but at a cost of increased blur.

I think they called it Quincunx, not because that's some esoteric term, but because that's what it sounds like when you vomit. Which is more or less what the screen looks like when that mode is enabled.

 

[Silent_One]

Having just purchased a 9700 Pro I must say I'm quite impressed with 6X FSAA. Having seen a V5 in action IMHO the 9700 is just as good.....

IF I recall awhile back Reverend put up screen shots of different FSAA methods of which the V5's were the best looking ones. Perhaps it's time for another review?

 

[Fafner]

I seem to remember a comparison between Voodoo5 5500 and Radeon8500 FSAA but cant find it now. Here's a link to a slightly biased comparison between a V5 and Gf3, its a fun read though, and the dude got some points.
http://www.papadoc.net/V5vsGF3.html

 

[renmike]

LOL...

that comparison above sucked....talk about BIASED...

check out the IL2 screenshots....the Nvidia image has tons of JPEG artifacts, the 3DfX one does not have any that I can notice...

also, he compared the GF3 to the V5 in 3DMark 2k1 in 4X-AA...but he crippled the GF3 by ALSO enabling 8x Aniso....at least the GF3 still won lol....

sheesh....

 

[martrox]

Anyone who has or has had a V5 can attest to the fact that nVidia FSAA on ANY nVidia card, including the TI4600 @4Xs cannot compare to the V5 4X..... of course, that's just talking FSAA!

 

[jb]

I had both and its easy to see. The GF4 drops back to an ordered grid so it looses some of its ability to remove jaggies on veritcal/horizontal edges. Also the 4xS mode helps on alpha and does a touch of texture filtering, the V5 does a better job at both. So if its in AA quailty only then no doubt V5. However since the V5 is so slow.....its not exactly fair

 

[Bambers]

The skewed grid that 4xS gave allowed the horizontal edges to be equal to a RGAA but the vertical edges were still poor. AA on alpha textures was effectively limited to the same as a 2x1 OGSS.

 

[Fafner]

well I said "slightly biased"...