PDA

View Full Version : When I enable 4xAA in a Game, is it actually rendering at 16AA on G80?


anticitizenzero
11-11-06, 09:06 PM
I read that the 16x AA on the 8800 GTX does it at the cost of 4x AA, so when I enable 4x on in game settings such asdoes the card render it as 16AA? I ask this because I tryed forcing 16xAA instead of enabling it in a game, and when I do performance benchmarks, there is no drop in FPS.

P.S I also noticed there is also there is 16xq antialiasing, what is the difference between that in 16x?

sillyeagle
11-12-06, 05:51 PM
You are partly correct. You still have to specify 16xAA in the CP to render at that setting. So you can either set the CP to override the application setting, then specify 16xAA in the CP, or you can set the CP to enhance the application, in which case you set the application to 4x or any setting for that matter, then you again specify 16xAA in the CP. Either way you have to specify 16xAA in the CP, or whatever you want.

The difference between 16x and 16xQ is that the Q stands for quality, and it gives huge performance hit, with little noticeable quality improvement.

I have found the settings you want to use are 4x, 8x, and 16x. 8xQ and 16xQ give bigger performance hits.

I created a thread a while back that details the difference between 16x and 16xQ, with full charts and benchmarks.

http://www.nvnews.net/vbulletin/showthread.php?t=79858

EDIT: In particular read posts #3 and #4.

Dazz
11-12-06, 05:59 PM
16x is very fast compaired to 16Q, with a 2.5GHz Athlon you wouldn't see the hit. People need 3+GHz core to processors and even then they are still GPU limited and the core 2 is around 20% faster then an Athlon. So to answer that no you wouldn't see a performance hit.

sillyeagle
11-12-06, 06:50 PM
...with a 2.5GHz Athlon you wouldn't see the hit. People need 3+GHz... So to answer that no you wouldn't see a performance hit.

Thats not entirely true....

I have a 2.7GHz and I certainly see a LARGE hit with 16xQ enabled over 16x in both Oblivion and FEAR.

In the scene below at 1600x1200 I get 34 FPS with 16x, and 26 FPS with 16xQ, and thats with over 100GB/s of memory bandwidth.

http://sio.midco.net/111lll/Oblivion.jpg

sillyeagle
11-12-06, 07:02 PM
I just tested FEAR and at 1600x1200 I get 75 FPS with 16x an 36 FPS with 16xQ.

So I would aviod 16xQ at all costs, as well as 8xQ, as they both store in memory a z/color value which is twice that of 4x/8x/16x. Even with 100GB/s it is too much. So 4x, 8x, 16x are the settings to go with.

Arioch
11-12-06, 07:05 PM
If someone could post some screenshots of the same scene with 16X and 16XQ so we can see the difference it would be appreciated.

sillyeagle
11-12-06, 08:01 PM
If someone could post some screenshots of the same scene with 16X and 16XQ so we can see the difference it would be appreciated.

Here you go. They are both 1600x1200 saved at maximum quality JPEG, so they are about 1MB each. For some reason some of the beam don't have AA applied very well for some reason. I'm using Multisample transperency AA if it makes any difference.

The first is 16x, the second is 16xQ, and the framerate is in the corner...

http://sio.midco.net/111lll/hl2.jpg

http://sio.midco.net/111lll/hl2b.jpg

Arioch
11-12-06, 08:15 PM
Thanks for posting those. Maybe I am getting old but I don't see enough of a difference to warrant the lower performance. As I plan on playing Oblivion again with my new PC I would like to see some shots of that game maybe.

knghtwhosaysni
11-12-06, 08:20 PM
I don't see any change either, but it looks like there are still quite a few jaggies

sillyeagle
11-12-06, 09:02 PM
I don't see any change either, but it looks like there are still quite a few jaggies

Yeah I see no difference in quality, only performance.

As to the jaggies, somebody else with more tech knowledge than me may know why, but certain surfaces don't get AA applied like the normal "objects" in the game. If you look at the main beam in the frame, you can see there are zero jaggies, but of course the cross beams and stuff above have them.

Xion X2
11-12-06, 09:05 PM
I think something funny's going on w/ mine.

On Oblivion, on a single 8800GTX, I'm getting as low as 30FPS in thick forests only running 4xAA @ 1680x1050 resolution. And when I up the AA to 16x I get nearly identical performance.

I hate this new control panel nvidia's come out with. I remember it doing some screwy stuff with the GX2 when they released it. I wonder if it's messing up the settings on G80 as well.

Absolution
11-12-06, 09:10 PM
Some beams are probably just textures thus they don't get AA applied unless you have TAA on (SS).

Same thing with the fence. Just a 2d texture so you need TAA.

sillyeagle
11-12-06, 09:12 PM
Here are a couple 400% crops to compare. the first one is 16x, the 2nd 16xQ.

http://sio.midco.net/111lll/zoom16x.jpg

http://sio.midco.net/111lll/zoom16xQ.jpg

CaptNKILL
11-12-06, 09:18 PM
Strange, you said you're using Transparency AA but it doesn't appear to be working on the fence or the higher beams in the ceiling (both are sprites).

Isn't that the whole point of TAA?

sillyeagle
11-12-06, 09:19 PM
Some beams are probably just textures thus they don't get AA applied unless you have TAA on (SS).

Same thing with the fence. Just a 2d texture so you need TAA.

Hmm thats strange, when I quote you I get a 2nd sentence which is not in your post.... I've even refreshed the page a few times and this 2nd sentence is not there, only when I quote...

Anyway, I do have Multisample TAA enabled in that scene. I can try Supersample and see if it makes a differece.

sillyeagle
11-12-06, 09:33 PM
Ok I fixed the jaggies on the upper beam with Supersample TAA, as it seem Multisample TAA is not enough to do it. Take a look at 16xQ, with Supersample TAA.

Unfortunatly look at the frame rate hit. From 60 FPS with multisample TAA to 30 FPS with Supersampe TAA.

http://sio.midco.net/111lll/hl2c.jpg

knghtwhosaysni
11-12-06, 09:37 PM
How about supersample with 16x?

It's seems odd to me that the framerate hit is so big, because on the 7 series it was like a 4fps difference between multisampling and supersampling transparency... maybe the hit gets exponentially bigger with these crazy-high AA modes like 16xQ

But I can tell that supersampling definitely helped :)

sillyeagle
11-12-06, 09:43 PM
How about supersample with 16x?



Yep this is the sweet spot. No jaggies whatsoever, and 70 FPS. I ran around this whole level and its in the 70-110 range, and thats at 1600x1200, 16x with Supersample TAA.

http://sio.midco.net/111lll/hl2d.jpg

CaptNKILL
11-12-06, 09:50 PM
Holy s**t batman... :p

This... is awesome. 16xSS TAA at 1600x1200 at 70-110fps?

I guess I can live with that... :D

sillyeagle
11-12-06, 10:10 PM
I tested Multisample vs Supersample TAA in Oblivion and unlike HL2 SS is too much.

So at 1600x1200 I can use 16xAA with SSTAA in HL2 Ep1, but in Oblivion to get good framerate at 1600x1200 I use 8xAA with MSTAA.