[CosmoKramer]

What's up with this question? Experiment and see for yourself what you prefer.

[Ailuros]

Originally Posted by Natoma

Hey Rev or Dave. Maybe you guys could include this in a review sometime, i.e. try to figure out a process that would qualitatively measure the effects of AA alone and AF alone on the image quality, much in the same way that the effects of 32bit color and Trilinear were measured.

Would that be possible? Or subjective as everyone seems to be saying?

I don't know if there's a method I'm not aware of, but usually it's hard to demonstrate texture aliasing for example, only describing it. Any antialiasing method that treats edges can be pictured with real game screenshots or via a testing application.

Reverend had a small article about negative LOD adjustments once and their side-effects; there using too aggressive negative LOD adjustments will create or increase (already existing) texture aliasing.

As for the original question of this thread it depends on way too many factors which have been mentioned already, like applications used, size of the monitor, antialiasing methods or even sampling grids etc etc.

IMHO those that have been and still have more a preference for Supersampling AA, can't mostly stand nasty side effects like texture aliasing, moire, pixel popping or any other possible noise in rendered scenes. There instead and in most cases a good MSAA sampling pattern and a fair amount of advanced texture filtering samples in combination with very high resolutions can eventually cure most problematic cases. Presupposition the accelerator can actually handle it.

Seeing the original poster considering an ultra high end card, the above combination shouldn't be an issue at all with most current games. In a worst case scenario where the game is already fillrate limited (which is rare actually) 6xAA is more than plenty for edges even in 1024*768, since the effective EER should be (if I'm not having a brainfart again) in the 6144*4608 league.

[BRiT]

For UT2003, I run at 1024x768 6xAA 8xAF on 9700Pro AMD ~3100+[11x200FSB], 1Gig memory, WinXP-SP1. I find that gives me extremely smooth and playable framerates even in intensive botmatches. When I use my LCD I run 1280x1024 4xAA 8xAF.

Now adays it shouldn't be a question of one or the other, but what balance between the two.

[Reverend]

Here is what I did many years ago to show off AA in motion in a V5 5500PCI review :

http://www.beyond3d.com/misc/reveren.../NFS5demo.html

Be patient however... for it to work, it needs to load 51 images totalling about 3.8MB

I'm only posting this because it seems there are still millions of folks out there that don't seem to give AA its due importance :? 8)

Gawd... Dave is gonna kill me for the bandwidth overload...

[vrecan]

to me it matters on the game, in a fast paced fps I dont see the need for AA I dont even notice it! But I do notice AF still so if performance is a prob first thing I get rid of is AA. In other games like flight sims AA makes an ugly thing a wonderful thing so it definetly has its place!

[Althornin]

Originally Posted by vrecan

to me it matters on the game, in a fast paced fps I dont see the need for AA I dont even notice it! But I do notice AF still so if performance is a prob first thing I get rid of is AA. In other games like flight sims AA makes an ugly thing a wonderful thing so it definetly has its place!

I always notice lack of AA and AF.
If you are moving fast, all that does is make the jaggies sparkle more.
No thanks.

[Seiko]

Well although many will probably disagree I honestly believe you could get away with a negative lod and high level of FSAA. This would give a very close comparison between high AF and high FSAA but would save on the AF penanlty. Although pixel shimmer can become a problem the high FSAA (i.e. 4*) on the V5 removed that quite nicely. I'm not sure that the R300/R350s FSAA would but I wish I had a lod bias to play with. Anyway, thankfully the 9700Pro makes mince meat of the two settings (for now at least) so I play UT 10*7, 6*FSAA, 16*AF.
Although having read the recent threads on AF panel settings application mode/quality mode etc I wonder what I've been getting anyway?

In short, FSAA is a must but so are clear [edit, sharp] (shimmer free) textures!


: )

[bloodbob]

AF all the way unless your running a really pathetic resolution ( or really big display ) if you got low res it prolly cause of preformance reasons and then u don't want AA/AF. I personally find 1 problems with the R300 and AF ( well on 16x atleast ) the LOD bias on mipmaps seems to be to high or something as I get alot of noise :/

[sonix666]

Well although many will probably disagree I honestly believe you could get away with a negative lod and high level of FSAA. This would give a very close comparison between high AF and high FSAA but would save on the AF penanlty.

I disagree, seeing the difference, negative LOD values are in no way a close substitution to AF.

[Colourless]

I'll take a moderate amount of FSAA and Aniso rather than take an all of one, nothing of the other approach. I'll use 4x FSAA and 8x Aniso if higher settings are slow.

But, if i was forced to choose between the two, I'd go for FSAA. I've been using it for far too long now to go back.

[Reverend]

Originally Posted by Colourless

But, if i was forced to choose between the two, I'd go for FSAA. I've been using it for far too long now to go back.

Is that because of FSAA per se or because the V5 6000, that you'd been using for so long, runs AA (whatever # of samples) at reasonable pace (according to your preference, thresholds and games you check out regularly)?

Most folks don't "like" FSAA because of the performance hit, which is more than simply upping the rez.

[Colourless]

Originally Posted by Reverend

Is that because of FSAA per se or because the V5 6000, that you'd been using for so long, runs AA (whatever # of samples) at reasonable pace (according to your preference, thresholds and games you check out regularly)?

Most folks don't "like" FSAA because of the performance hit, which is more than simply upping the rez.

Actually, it's yes on all counts. The V5 6000 ran 4x FSAA well, and looked great, so I'm spoiled there. Of course, I can't say i've ever required the fastest of fastest frame rates (I never play multiplayer competitively), and I'm often not playing the most recent games out there.

Then there are other hardware limitations. I'm using a 17" monitor and I can't run at 85Hz in any res above 1024x768, so that is what I play at. Then... then of course, there is the other one... for a vast majority of the time I'll probably be CPU limited, so the hit for FSAA is relatively small.

Now, if the question required that i have to use bilinear filtering with FSAA, vs trilinear anisotropic with no FSAA, then things would be different. I find the mipmap bands with standard bilinear filtering totally irritating.

[Malo]

Colourless, not actually getting into the UT2003 bi/tri debate are you?

[Bambers]

FSAA has less of a hit than upping the res on modern cards.

I've noticed that 1600x1200 4xAA is about the same speed as 2048x1536 no AA on my 9800pro, 2048 normally has higher maxes and lower mins though,

[Ailuros]

I'm not sure that the R300/R350s FSAA would but I wish I had a lod bias to play with.

What does the "mipmap detail level" in both APIs stand for?

But, if i was forced to choose between the two, I'd go for FSAA. I've been using it for far too long now to go back.

I figure by FSAA you actually mean Supersampling. IMHO you get Full Scene Antialiasing only once you actually combine MSAA with AF. In a case example where the game has f.e. problematic mipmapping and a AA/AF combination isn't possible due to performance restrictions, I personally would rather have just AF and skip the AA part. Edges are the smaller proportion of a given scene

Supersampling is a more "two in one" sollution compared to MS/AF, yet with recent fillrate (and partly bandwidth) saving sollutions of the latter, the influx on fillrate with SSAA is simply not worth it. Besides I don't think that Supersampling alone cures texture aliasing entirely in a problematic case, an oversampled SSAA+AF combination only would.

By the way the default LOD on the V5 was quite a bit lower than on other cards, even more the recent ones. LOD adjustments can be made on recent high end cards, I use a more conservative (positive) setting for both APIs most of the times.

Finally just using Multisampling with no AF is unacceptable IMO.

[Colourless]

Originally Posted by Ailuros

I figure by FSAA you actually mean Supersampling.

Y... No, not at all. If I intended to mean supersampling, i would have stated that.

[Natoma]

Originally Posted by Reverend

Here is what I did many years ago to show off AA in motion in a V5 5500PCI review :

http://www.beyond3d.com/misc/reveren.../NFS5demo.html

Be patient however... for it to work, it needs to load 51 images totalling about 3.8MB

I'm only posting this because it seems there are still millions of folks out there that don't seem to give AA its due importance :? 8)

Gawd... Dave is gonna kill me for the bandwidth overload...

That's a fantastic app there. Have you considered creating one that would show the various differences between AF and no AF in motion? Sort of like what MikeC did with his videos, but split screen them.

Thanks everyone. I've pretty much settled on 4xAA and 16xAF. In botmatches I get a little slow down every once in a while, but it's rare. My eyes literally popped out of my head when I saw the difference in UT2K3 between no AA/AF and 4xAA/16xAF. Lord. Like a whole new game.

[Ailuros]

Uh oh MSAA without AF?

Of course is it a pure personal preference, but for me the following is a nah uh:

http://users.otenet.gr/~ailuros/4MSAA.jpg

...while I could still live with that (just to illustrate the difference) in a worse case scenario:

http://users.otenet.gr/~ailuros/16AF.jpg

[Chalnoth]

Supersampling improved texture quality as well as edge quality.

Anisotropic filtering is best used in conjunction with MSAA, so that one can have better edge quality and better texture quality for the same performance.

MSAA offers better performance because the only significant hit is on memory bandwidth.

Anisotropic filtering offers better performance because it only adds texture samples for those pixels that need it (whereas SSAA is global).

MSAA and anisotropic filtering are best used together because MSAA primarily impacts memory bandwidth, while anisotropic filtering primarily impacts fillrate. This means that enabling one or the other alone will have less of a relative performance hit than enabling both together.

We should not have to choose between them. They affect different aspects of image quality, and should both be used.