Why AA doesn't get the respect it probably deserves

R

Reverend

Banned
We really have basically no control over this. As far as we're concerned, we have the survey about baseline machines and that is basicvally our target. Anti-aliasing is not our concern and I doubt it will ever be for us or any other developer. Our wish as developers is for gamers with our targetted baseline system to turn on all graphics details or shader effects (btw, you are going to like what we did) and have a near solid ~100 Hz. We mess around with configurations and at the lowest end we still want 45-60 Hz. AA never comes into the picture. AA is a needless tech support headache actually because as you know one of the first things that are recommended for gamers experiencing slow performance is to check their AA settings. To be honest, I wish we can have a sticker on every jewel case with the words "Don't fucking use AA". Sure, we're not talking about the 3D-savvy gamers... but these surveys don't lie.
Click to expand...

I am pretty sure some/many of you are probably familiar with the above scenario because that's actually what we're experiencing. AFAIK, I don't think there are any PC games that ship with a default graphics configuration that applies at least 2xAA. Obviously, there are so many types of PC systems out there to consider for one single game. Quite unlike the console platform.

What, in your opinion, would be required for the above to change? Yep, this really is a BIG open question but I would personally be interested in all kinds of inputs (except for a "Just tell all IHVs to discard any kind of AA"). There are a myriad of things to consider, for sure.
 
  • Like
Reactions: Geo
Totally agree with him to be honest, far far better for a gaming experience to actually have first good framerates and then good effects in the games. After that AF takes priority and then bottom of the pile comes AA.
 
dizietsma said:
Totally agree with him to be honest, far far better for a gaming experience to actually have first good framerates and then good effects in the games. After that AF takes priority and then bottom of the pile comes AA.
Click to expand...
I agree about AF being more important, but because it's so easy to implement (especially with control panel AF being almost bulletproof) I don't see why that should enter the equation.

However, I disagree with your reasoning. Performance is never a problem because the user can drop a resolution level easily to get a 60% boost in framerates (assuming you're pixel limited). The same thing applies to AF. I will always take a lower resolution with AF/AA. I don't even know why we test without AF (you'd be a fool to disable it), as it always make the AA hit seem larger.

Anyway, Rev's quote sort of lends credibility to my theory that the reason we don't see AF or AA on most console titles is apathy. Not enough console gamers care, so neither do the devs. They'd rather just cut the man-hours devoted to optimization.
 
Didn't FEAR apply 4xAA by default (via the simple options panel, as opposed to the detailed one)?

In that game at least, adding AA whilst using a lower resolution certainly did seem preferable, both for performance and image quality. I suppose this is the exception rather than the norm.

Could I ask where that quote came from? Sounds like an interesting interview.
 
Tradationally consoles are played on blurry TV screens so I can understand AA and AF taking a back seat however with the new gen of consoles it is becoming more important.
 
First of all, for AF, I fail to see how that makes sense if you have G80-like performance scaling. Honestly, who cares about a 2-3% performance hit? As for AA, I personally wish a good and very cheap option was available; 2x MSAA + 6x coverage samples might be very interesting.

However, I suspect too many pixels would be "missed" by the 2 real colour samples. I really cannot imagine any distribution whatsoever with only 2 samples that have good hit rates, unlike with 4 samples. Either you don't have the samples near the edges, or you are actually missing two edges. Ah well! For G80-level performance cards, I guess 16x CSAA is already that sweet spot, though.

I know some people are going to kill me for even considering this, but I would like a really hacky edge AA mode. I believe most of the inherent artifacts of this technique would not be more noticeable in practice than aliasing in a decent implementation. Simply have a single "real" sample for Z and test against it if the triangle collides with the pixel's "bounding box". If it passes the test, the Z is replaced and the two colours are merged as per the coverage percentage.

Would such a technique have a number of image quality problems? Yes. But it's also ridiculously simple, and allows for decent cost-less antialiasing. I'd be curious to see a software rasterizer implement it in a modern game to see how good/bad it looks...
 
A substantial number of gamers would not mind playing their games without AA. Trading it off with performance hit of the order it currently carries is unacceptable for many. A game that ships with AA might not score well across the spectrum..

My interpretation of current AA techniques in hardware {correct me if I'm wrong} is that of a restricted (adaptive) oversampling of pixels, something that will almost always hit performance - the idea being to sample only a minimum number of extra pixels. My guess is that optimization on this number will soon saturate, restricting the performance with those techniques..

An off the cuff idea could be to let the display handle AA (incorporate it some way).. but I'm not sure I'll see that happening in the near future..
 
anjulpa said:
A substantial number of gamers would not mind playing their games without AA. Trading it off with performance hit of the order it currently carries is unacceptable for many. A game that ships with AA might not score well across the spectrum..

My interpretation of current AA techniques in hardware {correct me if I'm wrong} is that of a restricted (adaptive) oversampling of pixels, something that will almost always hit performance - the idea being to sample only a minimum number of extra pixels. My guess is that optimization on this number will soon saturate, restricting the performance with those techniques..

An off the cuff idea could be to let the display handle AA (incorporate it some way).. but I'm not sure I'll see that happening in the near future..
Click to expand...

I take back my idea, just realized its stupidity. Sorry..
 
Arun Demeure said:
I know some people are going to kill me for even considering this, but I would like a really hacky edge AA mode. I believe most of the inherent artifacts of this technique would not be more noticeable in practice than aliasing in a decent implementation. Simply have a single "real" sample for Z and test against it if the triangle collides with the pixel's "bounding box". If it passes the test, the Z is replaced and the two colours are merged as per the coverage percentage.

Would such a technique have a number of image quality problems? Yes. But it's also ridiculously simple, and allows for decent cost-less antialiasing. I'd be curious to see a software rasterizer implement it in a modern game to see how good/bad it looks...
Click to expand...

Well consider me first in that line of your potential murderers LOL :D

I've no single interest for half-assed wannabe features and that includes stuff like motion trail or other similar nonsense. Wherever I bump on to any of it my first thought is always "where the hell can I disable that crap".

I'm sure all major IHVs have spent quite a few manhours on research for that specific field and you know more than well how quickly any at first interesting appearing idea reaches the trashcan if it shouldn't work in only a couple of corner cases.

What, in your opinion, would be required for the above to change? Yep, this really is a BIG open question but I would personally be interested in all kinds of inputs (except for a "Just tell all IHVs to discard any kind of AA"). There are a myriad of things to consider, for sure.
Click to expand...

Would be forcing 2xAA from the IHVs side on all future GPUs (=/>2010) be acceptable input?
 
Frame-rate should come first, but anti-aliasing is important, I wish every game could look as good as 720x576 PAL DVD movies. The scene does not even have to photo-realistic, imagine a Southpark frame at DVD resolution with and without anti-aliasing. You could have the best shaders and the textures in the world, but without true multi-sampling, there's a strict limit to the belieavability of a scene. 1024x768+AA looks better than 1280x960 without AA etc etc.
 
Arun Demeure said:
Simply have a single "real" sample for Z and test against it if the triangle collides with the pixel's "bounding box". If it passes the test, the Z is replaced and the two colours are merged as per the coverage percentage.
Click to expand...
You really don't want order dependent opaque rendering, do you? :)
I also have the impression that even a few 'wrong' pixels randomily popping all over the image are not good (tm).
I'd really wish there were a way to do cheap AA..but I think such a thing just not exists and never will.
edit: I will spare your life for this time (j/k) :)
 
With all the great threads around this site, it's interesting that this is the one that finally got me to register after visiting the site for well on 3 years now.

Personally, I think AA should be the number 1 priority. AF should be the number 2 priority. Framerate should be the number 3 priority.

If a game cannot or will not support AA, I will not play it. End of story. And this considering I'm a FPS, RTS, and RPG junkie.

Nothing will turn me off to a game faster than seeing Jaggies everywhere I look. Nothing will scream crappy visuals more than Jaggies crawling all over whenever I move.

It's why the Voodoo 5 was my main video card until the GF4 was release even though I owned all versions of the Geforce series.

And until they have displays with extremely high DPI, this will continue to be and always will be problem. If a 24" display ever came out with 3840x2400 resolution I may revise my opinion, however for now. If a game doesn't allow the use of AA. It goes into the trash bin pronto.

Regards,
SB
 
i still think theres no reason devs shouldnt code their games so that control panel aa works. they have been doing so for years. games like r6 vegas, stalker etc rly frustrate me with their lack of aa support.
 
IMO the relative impact of AA and AF depends on the title - there's no universal rule that says AA > AF or vice versa.
 
You must log in or register to reply here.

Similar threads

R
Gaming and gaming hardware trends
2
Replies
23
Views
931
Remij
R
S
Harold Halibut [PS, XB, PC, XGP]
Replies
6
Views
689
BoardBonobo
BoardBonobo
R
Is there a way we can get Square Enix to sit up, pay attention, and fix some of their PC ports?
Replies
0
Views
378
Remij
R
Shortbread
Phil Spencer Interview: Redfall Reviews, Activision Deal - Kinda Funny Xcast Ep. 137
7 8 9
Replies
175
Views
9K
trinibwoy
T
S
Discussion about lighting (usage of lighting techniques).
2
Replies
21
Views
1K
Frenetic Pony
F
Back
Top