Far Cry 2 performance gains using AA

[Hauk]

Just ran the ranch small benchmark and realized a nice jump using AA.


Benchies @ 1280 x 1024

No AA = 68.20avg 100.66max 46.10min

2x AA = 75.36avg 115.71max 51.47min

4x AA = 73.92avg 111.26max 50.22min

8x AA = 62.67avg 89.48max 45.91min


Benchies @ 1920 x 1200

No AA = 50.64avg 68.33max 38.21min

2x AA = 73.27avg 111.72max 50.09min

4x AA = 72.51avg 110.41max 49.67min

8x AA = 43.07avg 63.71max 29.05min

As you can see, 2x and 4x AA saw nice gains, while 8x AA choked a bit.

*EDIT*
The gains stand after a couple experiments. Thanks to those who've helped.

Settings
DX 10
All in game settings maxed
VSync disabled
Latest Catalyst 8.10 hotfix drivers

Rig
Vista 64 Premium
Intel E8600 @ 4.2
Asus Rampage Formula
8GB Muskin Redline DDR2 1000 @ 1019 5-5-5-12
4870 X3 (4870 X2 @ 770core 925mem + 4870 1GB @ 770core 925mem)
WD Velociraptor 300GB

 

[dguy6789]

Would someone care to explain to me why enabling AA would improve performance in any scenario at all? I've never heard of this.

 

[SSChevy2001]

Here's 1 run @ 1080p maxed on my 4870

52.43avg 72.70max 39.77min

Comparing the scores it looks like only one GPU is working on your setup without AA.

 

[error8]

Originally posted by: SteelSix
What's the deal with 8x AA choking? T

You run out of video memory at 8xAA. You're triple 4870 has only 1 gb of memory basically, those other 2 gb are "mirrored", they have the same data in them as the other gb has, so it seem that Farcry 2 needed more ram to run 8X AA better.

 

[Hauk]

Originally posted by: error8

Originally posted by: SteelSix
What's the deal with 8x AA choking? T

You run out of video memory at 8xAA. You're triple 4870 has only 1 gb of memory basically, those other 2 gb are "mirrored", they have the same data in them as the other gb has, so it seem that Farcry 2 needed more ram to run 8X AA better.

I just ran 1280 x 1024 and saw the same thing with 8x AA. Guess even with lower resolution it's chewing up memory?

 

[Hauk]

Originally posted by: SSChevy2001
Here's 1 run @ 1080p maxed on my 4870

52.43avg 72.70max 39.77min

Comparing the scores it looks like only one GPU is working on your setup without AA.

Hmm, I'm sure scaling could be better but only one GPU without AA? Not sure about that. Probably worth a couple people testing no AA vs. 2x and 4x AA..

 

[error8]

Originally posted by: SteelSix

Originally posted by: error8

Originally posted by: SteelSix
What's the deal with 8x AA choking? T

You run out of video memory at 8xAA. You're triple 4870 has only 1 gb of memory basically, those other 2 gb are "mirrored", they have the same data in them as the other gb has, so it seem that Farcry 2 needed more ram to run 8X AA better.

I just ran 1280 x 1024 and saw the same thing with 8x AA. Guess even with lower resolution it's chewing up memory?

Well it probably does, since you're seeing that drop in performance.

 

[SSChevy2001]

Originally posted by: SteelSix

Originally posted by: SSChevy2001
Here's 1 run @ 1080p maxed on my 4870

52.43avg 72.70max 39.77min

Comparing the scores it looks like only one GPU is working on your setup without AA.

Hmm, I'm sure scaling could be better but only one GPU without AA? Not sure about that. Probably worth a couple people testing no AA vs. 2x and 4x AA..

Here's 12x10 without AA on single card.

62.18 avg 82.73max 47.83min

 

[Hauk]

Originally posted by: SSChevy2001

Originally posted by: SteelSix

Originally posted by: SSChevy2001
Here's 1 run @ 1080p maxed on my 4870

52.43avg 72.70max 39.77min

Comparing the scores it looks like only one GPU is working on your setup without AA.

Hmm, I'm sure scaling could be better but only one GPU without AA? Not sure about that. Probably worth a couple people testing no AA vs. 2x and 4x AA..

Here's 12x10 without AA on single card.

62.18 avg 82.73max 47.83min

Forget single vs. dual vs. tri GPU scores for a second. You've proven once before that scaling (in general) sucks.

Try no AA vs. AA at 2, 4, 8x

 

[YBS1]

How strange. I tried it after seeing this post and I'm getting a similar result. I'm only running a single 4850 so it did choke on 1920x1200 with 4X AA, but my 1920x1200 2X AA bench was indeed faster than that of my no AA bench. Settings at ultra quality DX10. It would be nice if ATI would sort out the DX10 hitching issue as the game does run quite a bit faster under DX10 as opposed to DX9.

Nevermind the above. I went back and ran the benchmarks over using a three run loop instead of just a quick single run and they turned out differently.

1920x1200 Ultra Quality No AA
* Average Framerate: 29.37
* Max. Framerate: 37.04
* Min. Framerate: 24.72

1920x1200 Ultra Quality 2X AA
* Average Framerate: 21.91
* Max. Framerate: 30.57
* Min. Framerate: 15.35

1920x1200 Very High Quality No AA
* Average Framerate: 46.91
* Max. Framerate: 65.06
* Min. Framerate: 34.13

1920x1200 Very High Quality 2X AA
* Average Framerate: 40.73
* Max. Framerate: 57.05
* Min. Framerate: 30.72

 

[apoppin]

Your demo runs are so far apart as to be meaningless

Disable AI and remove the CPU limitation and try again

 

[SSChevy2001]

Originally posted by: SteelSix

Originally posted by: SSChevy2001

Originally posted by: SteelSix

Originally posted by: SSChevy2001
Here's 1 run @ 1080p maxed on my 4870

52.43avg 72.70max 39.77min

Comparing the scores it looks like only one GPU is working on your setup without AA.

Hmm, I'm sure scaling could be better but only one GPU without AA? Not sure about that. Probably worth a couple people testing no AA vs. 2x and 4x AA..

Here's 12x10 without AA on single card.

62.18 avg 82.73max 47.83min

Forget single vs. dual vs. tri GPU scores for a second. You've proven once before that scaling (in general) sucks.

Try no AA vs. AA at 2, 4, 8x

Not sure why I did all these test, but again to me looks like your not scaling without AA. Clearly I take a hit when enabling AA.

12x10 0xAA Ultra 62.18avg 82.73max 47.83min
12x10 2xAA Ultra 58.02avg 77.11max 41.12min
12x10 4xAA Ultra 52.02avg 71.81max 38.26min
12x10 8xAA Ultra 38.65avg 59.11max 24.75min

19x10 0xAA Ultra 52.43avg 72.70max 39.77min
19x10 2xAA Ultra 46.79avg 65.56max 35.05min
19x10 4xAA Ultra 41.11avg 64.93max 29.49min
19x10 8xAA Ultra 27.89avg 65.12max 17.58min

 

[Hauk]

Originally posted by: apoppin
Your demo runs are so far apart as to be meaningless

Disable AI and remove the CPU limitation and try again

I just realized you were probably talking about disabling artificial intelligence in the benchmark program. I thought you were talking about Cat. A.I. so I disabled and wow, that made a difference. Scores dropped accordingly.

I re-ran benchies with Cat A.I. enabled and confirmed the gains with 2x AA and 4x AA. Whatever it is, it's tied to the optimizations in Cat A.I.

 

[apoppin]

Originally posted by: SteelSix

Originally posted by: apoppin
Your demo runs are so far apart as to be meaningless

Disable AI and remove the CPU limitation and try again

I just realized you were probably talking about disabling artificial intelligence in the benchmark program. I disabled Cat. A.I. and wow, that made a difference. Scores dropped accordingly. I re-ran benchies with Cat A.I. enabled and confirmed the gains with 2x AA and 4x AA. Whatever it is, it's tied to the optimizations in Cat A.I.

Don't disable Cat AI

look at what happens in CCC .. Xfire stops working and it really messes with an X2 also

Disable Artificial Intelligence in the game benchmark. You are getting the CPU involved and all your benchmarks runs are like 10 FPS apart at times; the average is one of extremes
--that means *nothing*

all 3 should be close to identical .. maybe an single FPS or two difference

 

[Hauk]

Originally posted by: apoppin

Originally posted by: SteelSix

Originally posted by: apoppin
Your demo runs are so far apart as to be meaningless

Disable AI and remove the CPU limitation and try again

I just realized you were probably talking about disabling artificial intelligence in the benchmark program. I disabled Cat. A.I. and wow, that made a difference. Scores dropped accordingly. I re-ran benchies with Cat A.I. enabled and confirmed the gains with 2x AA and 4x AA. Whatever it is, it's tied to the optimizations in Cat A.I.

Don't disable Cat AI

look at what happens in CCC .. Xfire stops working and it really messes with an X2 also

Disable Artificial Intelligence in the game benchmark. You are getting the CPU involved and all your benchmarks runs are like 10 FPS apart at times; the average is one of extremes
--that means *nothing*

all 3 should be close to identical .. maybe an single FPS or two difference

I normally run Cat A.I. enabled but thanks for the tip. Same results disabling artifical intelligence:

No AA No artificial intelligence = 52.96avg 70.12max 41.62min

2x AA No artificial intelligence = 82.84avg 126.81max 55.09min

4x AA No artificial intelligence = 69.59avg 95.54max 49.72min

I can't imagine why I'm the only one seeing this. I haven't smoked anything today I promise...

 

[apoppin]

Well all of your runs should closely "match"

if they do not, it is usually the game AI

most of my runs are really really close

if they are not, i just keep running them
[after exiting the bench; even to restarting Vista and then waiting the necessary time to begin benching]
these are YOURS - i bolded the inconsistency:

Loop 1 No AA:
Avs - 56-48-55 = average 53
Max - 74-68-74 = average 70
Min - 43-33-43 = average 43

not so close - 6 to 10 fps difference between 3 runs - i'd hate to pick a video card based on this test - up to 25% margin of error


Loop 2 2x AA:
Avs - 85-85-82 = average 83
Max - 133-125-126 = average 127
Min - 57-54-55 = average 55

very close comparatively and more like my figures match each other - 8 fps difference within 3 runs is not so severe out of 125 fps; although i would prefer half that


Loop 2 4x AA:
Avs - 62-67-80 = average 70
Max - 93-105-124 = average 95
Min - 33-47-50 = average 50

93 to 124 is ridiculous; 33 to 50 is a huge margin of error .. like + or - 33% off



look at the variability among your runs; it is not supposed to be that way

 

[Hauk]

I appreciate the help and info apoppin, you obviously know what you're doing. But it seems like you're splitting hairs here. Not sure what you're getting at with this variance thing, that my hardware is flawed? That my testing methed is flawed?

I don't have a freakin lab set up here and don't claim to be a professional benchmarker, but I don't see how I could screw up something as simple as lauching a built-in benchmark program. It's a fresh OS install, latest drivers, etc. It's kinda hard to screw this up.

Ran ranch small again, 2 and 4x AA are still faster than with no AA. Ran ranch demo playback, same results:

Playback Ranch Demo No AA = 56.95avg 76.15max 40.66min

Playback Ranch Demo 2x AA = 82.10avg 106.76max 50.56min

 

[Tempered81]

So I take it you're running 4xAA from here on out? heheh.

Steelsix, how do you think your rig compares to i7/x58/4870 crossfire?
http://www.guru3d.com/article/...-performance-review/10

In Far Cry 2 we selected an image quality setting, applied similarly to all cards. Since the graphics engine can deal with a lot, we opt for some of the very best image quality available in-game, DirectX 10 mode and then enable 4 levels of anti-aliasing in very high image quality settings mode. A pretty tough configuration for sure. The result is excellent image quality, yet it remains quite playable throughout the scope of tested cards and resolutions.

Image Quality Settings:

* 4x Anti Aliasing
* Textures: Very High
* Shadows: Very High
* Overall Quality: Very High
* DirectX 10

 

[Hauk]

You bet jared, 4x AA all the way. :beer:

Hey that's a good review you found. Are you trying to break me??
Damn what a difference the new platform and i7 makes for multi-gpu

 

[Tempered81]

No, just wondering if you had some 4870 X-fire, or single 4870X2 benches of Farcry2. I was curious about the performance of a 4.2ghz E8xxx/ X48 vs. 3.8ghz corei7/ X58, and your setup vs. guru3d is a nice comparison.

:beer:

 

[Tempered81]

Originally posted by: SteelSix
Benchies @ 1920 x 1200

4x AA = 72.51avg 110.41max 49.67min

Vista 64 Premium
Intel E8600 @ 4.2
Asus Rampage Formula
8GB Muskin Redline DDR2 1000 @ 1019 5-5-5-12
4870 X3 (4870 X2 @ 770core 925mem + 4870 1GB @ 770core 925mem)
WD Velociraptor 300GB

So far this bench ^^ compares to their i7/X58 testing.

They compared:
Two 512mb 4870's (2 gpus) at 1920 x 1200 4xAA and got 63 FPS average.
While 4870X2 quadfire (4 gpus) at 1920 x 1200 4xAA had 94 FPS average.
3 gpus should come in around ~ 78.5 FPS.
http://www.guru3d.com/article/...-performance-review/10

Pretty comparable to your 72.5 FPS. How man frames are you getting in this bench, apoppin?