GTX680 vs HD7970 with SGSSAA, user input requested

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

DimmyK

Member
Oct 26, 2010
137
0
86
Thx! Could you benchmark Heaven 3.0, too? Or what other games do you have? :)

Lets stick to 1080p for all future benchmarks, makes it easier for me to put it all together in a table. Thanks in advance.

Sure, I'll do heaven when I get a chance. I have bunch of games, just let me know which ones you want me to bench and I will if I have them.

Edit: I have Hitman Absolution installed currently, can do that after Heaven.
 

DimmyK

Member
Oct 26, 2010
137
0
86
GTX 680 @ 1280/3335, 310.61

Heaven 3.0

Settings:
Heaven%203.0%20settings.png


4xMSAA

Min: 14.9
Max: 181.0
Avg: 81.8

4xSGSSAA

Min: 25.9
Max: 112.1
Avg: 44.4


Hitman Absolution

Settings:
Hitman%20settings.png


4xMSAA

Min: 50.0
Max: 79.6
Avg: 62.3

4xSGSSAA

Min: 35.8
Max: 95.3
Avg: 45.6
 

Granseth

Senior member
May 6, 2009
258
0
71
7970 925/1375
Shogun 2
Every setting to max(Ultra)
Every effect activated (like HDR, Tessellation, Ambient Occlusion etc.)
1920x1200

4xMSAA
Average FPS: 54.2

4xSSAA
Average FPS: 47.1

Maybe I'm CPU bound or something, because that is not a big dip in FPS.
CPU is 3570K at 4.2 GHz, and have 16GB of memory.
 

Alusan

Member
Mar 5, 2010
33
0
66
Would you mind if I put up Black Mesa benchmarks for comparison later today? I'm interested to see if Tahiti loses a lot more performance than GK-104 for all Source Engine games like the Anandtech results for Portal 2 show, or maybe it's just the case with the newer build Portal 2 uses.

I think the tram ride at the very beginning works best for a benchmark since it's pretty static but you still go through several areas.
 

DimmyK

Member
Oct 26, 2010
137
0
86
GTX 680 @ 1280/3335, 310.61

Metro 2033

Settings:
Metro%20Settings.png


2 runs, 2nd run results:

4xMSAA

Avg: 73.77
Max: 137.87
Min: 18.14

4xSGSSAA

Avg: 42.98
Max: 79.57
Min: 13.16
 

cmdrdredd

Lifer
Dec 12, 2001
27,052
357
126
Heaven 3.0 16xAF, tessellation normal

GTX 670 SLI from sig on driver version 310.61

1920x1080

4x MSAA
min - 68.7
avg - 148.4
max - 321.5

4x SGSSAA
min - 42.3
avg - 72.3
max - 168.2

2560x1440

4x MSAA
min - 49.6
avg - 97.8
max - 224.8

4x SGSSAA
min - 26.8
avg - 47.3
max - 117.8
 

DimmyK

Member
Oct 26, 2010
137
0
86
GTX 680 @ 1280/3335, 310.61

Dirt Showdown

1080p, Ultra preset.

4xMSAA

Min: 56.01
Avg: 71.03

4xSGSSAA

Min: 32.83
Avg: 43.02



Shogun 2

Settings:
Shogun%202%20Settings.png


4xMSAA

Avg: 69.4

4xSGSSAA

Avg: 50.5
 
Last edited:

Gryz

Golden Member
Aug 28, 2010
1,551
204
106
Unigine Heaven:

4xMSAA:
---
Average FPS: 95.3
Min FPS: 14.5
Max FPS: 228.9

2nd run: 95.1 - 31.4 - 228.0

4xSGSSAA:
---
Average FPS: 37.9
Min FPS: 22.2
Max FPS: 97.4

It turned out I was testing the wrong settings at first.
These results were obtained later, when I figured out how to force 4xSGAA.

Btw, Min FPS is a useless stat. It is set during the first few seconds, and then never changes again. Seems like a crapshoot what you'll get.

1920x1080
gtx680 (@stock) 3570k (@4.0GHz)
Driver 310.64
16xAF, trilinear, tesselation normal.
shaders & textures high.
occlusion, refraction & volumetric enabled.
 
Last edited:

cmdrdredd

Lifer
Dec 12, 2001
27,052
357
126
Unigine Heaven:

4xMSAA:
---
Average FPS: 95.3
Min FPS: 14.5
Max FPS: 228.9

4xSGSSAA:
---
Average FPS: 91.9
Min FPS: 32.7
Max FPS: 225.4

1920x1080
gtx680 (@stock) 3570k (@4.0GHz)
16xAF, trilinear, tesselation normal.
shaders & textures high.
occlusion, refraction & volumetric enabled.

How did you get those numbers? Cause with SLI I'm not even at 90fps averages with SGSSAA. Also people with overclocked 680s are getting less than half what you claim.

Remember to use SGSSAA you need to enable MSAA to the same level since it uses MSAA to obtain sample coordinates. Maybe retest and make sure?

4x SGSSAA should also have 4x MSAA enabled.
 
Last edited:

lehtv

Elite Member
Dec 8, 2010
11,897
74
91
Not only are the SGSSAA numbers way too high, but the framerates for the 4x MSAA test look abnormally high as well. DimmyK got 81.8 fps avg with an overclocked 680, so 95.4 fps is definitely not possible with a stock 680 at the same settings. Perhaps something is overriding his AA setting
 

DimmyK

Member
Oct 26, 2010
137
0
86
Unigine Heaven:

4xMSAA:
---
Average FPS: 95.3
Min FPS: 14.5
Max FPS: 228.9

2nd run: 95.1 - 31.4 - 228.0

4xSGSSAA:
---
Average FPS: 91.9
Min FPS: 32.7
Max FPS: 225.4

2nd run: 92.4 - 37.8 - 277.3.

No AA:
---
Average FPS: 94.9
Min FPS: 36.0
Max FPS: 227.3

Very weird. It seems 4xMSAA hardly has a performance impact over no AA. Tested several times. And 4xSGSSAA is only a little slower than 4xMSAA ? I can also hardly see any quality difference between the modes. I had expected to see more jaggies when running without AA. But because the camera is constantly moving, I hardly notice them. There are some jaggies (like on the zeppelin), but I know that AA modes also sometimes miss jaggies.
It really makes me wonder what's going on.
Maybe I should retry these tests with Skyrim (looking at fps at certain fixed spots, straight from savefiles).

Btw, Min FPS is a useless stat. It is set during the first few seconds, and then never changes again. Seems like a crapshoot what you'll get.

1920x1080
gtx680 (@stock) 3570k (@4.0GHz)
Driver 310.64
16xAF, trilinear, tesselation normal.
shaders & textures high.
occlusion, refraction & volumetric enabled.

Definitely doesn't look right, it's almost like you're overriding AA settings in nvidia control panel (or in inspector). Sgssaa is not being applied for sure. Can you post screenshot of your control panel's default 3d settings profile?
 

Gryz

Golden Member
Aug 28, 2010
1,551
204
106
I had set 4xMSAA and 4xSGAA in the nVidia Inspector. With the "override the application's setting" option. I had expected this would override the application's setting, and make the application's setting not matter at all.

That was wrong.
I had to set the application from 0xMSAA to 4xMSAA.
Then the framerates dropped, and SGSSAA kicked in.
 
Last edited:

DimmyK

Member
Oct 26, 2010
137
0
86
Just a sec. I always assumed that the nVidia Control panel and the Inspector changed the exact same settings, right ? So it is impossible for the control panel to override the inspector ? E.g. when I configure 4xSGSSAA in the Inspector, the settings in the control panel change to "custom".

No, I'm pretty sure NCP can't override inspector changes, they change the same things, Inspector just way more granular. I think your heaven profile is the problem. At first, I wanted to do it via profile too, but then I saw like bunch of heaven profiles with no specific 3.0 profile. So I just enabled SGSSAA in default profile just for this run. Like this:

inspector%20sgssaa.png


I noticed that for some 3D applications, selecting just transparency supersampling in inspector won't apply SGSSAA, even if 4xMSAA is set in application. I need to specifically set "override" and MSAA settings in inspector too.
 

Gryz

Golden Member
Aug 28, 2010
1,551
204
106
Problem fixed now.
I edited my posts to reflect that.
 
Last edited:

lehtv

Elite Member
Dec 8, 2010
11,897
74
91
For the 4xAA test, just leave everything to application controlled and set 4x MSAA in Unigine Heaven. In the SGSSAA test, you still shouldn't need to override or enhance the application setting. Just enable 4x SGSSAA
 

DimmyK

Member
Oct 26, 2010
137
0
86
Ah, that might be it.
I had just prepared screenshots of my settings. :)
You can see 4 screenshots with settings here: http://hhwsmit.home.xs4all.nl/unigine/

Note, I ran all my tests with tesselation set to normal.
In an early test, I noticed that heavier tesselation settings give a pretty large impact on framerates.

Will try changing the global settings now. (Man, I hate that. :p)
I see you use "enhance the application setting" and texture filtering set to "quality".

I just tried heaven SGSSAA via profile and it works for me. "Enhance app settings" was your problem. It should be either mode/settings as "application controlled" and transparency supersampling to 4xSGSS or mode - "override", settings - "4xMSAA" and transparency SS to 4xSGSS.

This works, I can definitely see SGSSAA applied.
inspector%20heaven%20profile.png


I see you use "enhance the application setting" and texture filtering set to "quality".

I didn't use "enhance". In global profile, every setting was at default except AA mode - "override", AA settings - "4xMSAA", and AA - transparency supersampling - "4xSGSS". "Quality" is default value for AF.
 
Last edited:

Gryz

Golden Member
Aug 28, 2010
1,551
204
106
Messing with the Inspector profiles did nothing.
It turns out that with "override" or "enhance the application's settings", the in-game settings still matter. Tested with unigine and Skryrim. I suspect this is a new bug in 310.64 (or recent drivers). As I'm pretty sure I did see the framerate drop in Skyrim earlier, with in-game 0xMSAA.

BTW, I think I actually prefer the looks of 8xMSAA + SMAA over 4xSGSSAA in my games. SMAA removes a lot of the temporal aliasing. And 8xMSAA+SMAA has a higher framerate than 4xSGSSAA, making the game look more smooth. The textures might look a bit more blurry with SMAA, but I need to compare screenshots to make 100% sure.
 
Last edited:

MTDEW

Diamond Member
Oct 31, 1999
4,284
37
91
This pictures shows OSGSSAA and RSGSAA not SGSSAA.

OGSSAA = SSAA = downsampling entire image
RGSSAA = Rotated Grid Super-sampling Anti-aliasing:

Similar to SSAA. The difference is that instead of the sample grid being along horizontal and vertical axis, the grid is rotated to a fixed angle. This causes better anti-aliasing of edges that are almost horizontal or vertical.

SGSSA = Sparse Grid Super-sampling Anti-aliasing

Similar to SSAA. This type of AA is the first in a series of AA types that represent a compromise designed to trade image quality for performance.

The sampling pattern of plain SSAA is called Regular Grid or Ordered Grid. It can be pictured as a grid that has higher resolution than the pixel grid. Where SSAA samples in each square in that grid, SGSSAA samples in only some of the squares.

http://www.dahlsys.com/misc/antialias/index.html
First: There is no "SSAA". This term is incomplete. The first two characters, that denote the sample arrangement, are missing. It has to be either OGSSAA, RGSSAA or SGSSAA.
You said, Radeons don't do SGSSAA, but "traditional SSAA, which is heavier". That is not correct. They certainly don't do OGSSAA. Most of the time, OGSSAA does not trade image quality for performance. OGSSAA is a brute force method and looks worse than SGSSAA or RGSSAA at comparable performance.

I selected SGSSAA on my 580 to obtain the second picture. If it is SG or RG depends on how you want to look at it. It can be seen as rotated, but also as sparsed.

GTX580@stock
Heaven 3.0

DiogoDX, is correct.
AMD's SSAA is not the same as Nvidia's SGSSA.
 
Last edited:

DimmyK

Member
Oct 26, 2010
137
0
86
I got it working now.
I changed the global settings. Did a test. Still didn't work. I had high fps.
Then I deleted all the unigine profiles in the Inspector. Deleted half of them, could not delete the others, because Inspector would crash.

Then I wanted to run another test.
But I changed something else. I changed the setting in the unigine demo startup window from 0xMSAA to 4xMSAA.
Then my framerate dropped to half.

I changed the setting back to 0xMSAA. Framerates were back up.

So it seems you need to:
1) change settings in Inspector to 4xSGSSA and 4xMSAA
2) change settings in unigine-startup-window to 4xMSAA too

Result:
Average FPS: 37.9
Min FPS: 22.2
Max FPS: 97.4

I'm sorry for the confusion.
But when I configure something in the Nvidia Inspector, and configure "override application settings", then I expect to override the application settings. :/

You shouldn't be jumping thru all these hoops, and with Inspector crashing, this doesn't sound very good, so this is what I would do:

- Go to Nvidia control panel -> "Manage 3D settings" -> "Global settings" -> "Restore" -> "Apply"
- Uninstall Inspector and reboot
- Reinstall Inspector.
- Run Heaven without touching NCP or Inspector settings. Set all Heaven settings as in your previous run + 4xMSAA.
- In inspector, find "Unigine: Heaven demo" profile.
- Change 1 option: AA Transparency Supersampling to "4xSGSS"
- Run Heaven again with (same settings, 4xMSAA set in Heaven, not in Inspector).
 

boxleitnerb

Platinum Member
Nov 1, 2011
2,605
6
81
DiogoDX, is correct.
AMD's SSAA is not the same as Nvidia's SGSSA.

Source, proof? :)

Great turnout guys, I need some time to go through the results.
As for overriding AA - sometimes you need the compatibility bits for it to work. In DX10/11 you only need to set the 4xSGSSAA in transparency AA and that's not, nothing else is required (as long as 4xMSAA is selected in the game) :)

If you're GPU bound, 4xMSAA -> 4xSGSSAA should cost around 50% performance in most cases.
 
Last edited:

MTDEW

Diamond Member
Oct 31, 1999
4,284
37
91
Source, proof? :)

Great turnout guys, I need some time to go through the results.
As for overriding AA - sometimes you need the compatibility bits for it to work. In DX10/11 you only need to set the 4xSGSSAA in transparency AA and that's not, nothing else is required (as long as 4xMSAA is selected in the game) :)

If you're GPU bound, 4xMSAA -> 4xSGSSAA should cost around 50% performance in most cases.
The link's and info DiogoDX, provides gives the correct explanation as to why.
And having both Nvidia and AMD cards here, i can honestly tell you that AMD's SSAA provides a slightly cleaner look than Nvidia's SGSAA albeit with a larger performance hit also.

The conclusions you are reaching are based on testing only on Nvidia hardware , so its easy to see where you get the conclusions from.
And i am in no way implying you don't feel your conclusions are 100% correct.

But when you test them side by side, it is obvious that AMD's SSAA goes for the highest image quality regardless of the larger performance hit.

Whereas Nvidia's SGSSAA strikes the best balance between performance and image quality.
I hate calling it a "balance" because quite honestly Nvidia has darn near the same image quality and way better performance than AMD when it comes to AMD's SSAA vs Nvidia's SGSAA.

In other words, AMD's SSAA method seems more unrefined and a brute force method that simply isn't as optimized as Nvidia's SGSSAA.
As is clearly explained in DiogoDX's posts.
Whereas Nvidias SGSSAA seems much more refined and offers darn near indistinguishable results with way better performance than AMD's method.
The image quality difference is really only distinguishable in a side by side comparison.
And yet the performance result differences are easily distinguished by a FPS hit comparison. (Nvidia is just simply killing AMD here due to better optimization/options)

Seriously, i'm in no way trying to defend AMD's SSAA methods.
Quite the opposite, AMD needs to look at what Nvidia is doing with their SSAA options and get their own "optimizations" refined.

DiogoDX's,
Posts in this thread are accurate and shouldn't be so easily ignored if you haven't had the opportunity to test both AMD and Nvidia hardware side by side.
Yet you continuously ignore his posts and links/info.
example: SSAA samples in each square in that grid, SGSSAA samples in only some of the squares.
You continually keep saying that pure SSAA doesnt exist when it clearly Does exist and is what AMD's SSAA settings uses.

Super-sampling Anti-aliasing

This is the type of AA that creates the best image quality. It is “pure” AA with no compromises. All the other AA technologies exist because SSAA is very slow.

SSAA works by first calculating more pixels than what the resolution is on the screen, and then averaging those pixels to determine the colors of the actual pixels on the screen. The ratio of calculated pixels to displayed pixels is often described by a number put before the acronym. For instance, 8xSSAA describes that each pixel on screen will be the avarage of 8 calculated pixels.

Sparse Grid Super-sampling Anti-aliasing

Similar to SSAA. This type of AA is the first in a series of AA types that represent a compromise designed to trade image quality for performance.

The sampling pattern of plain SSAA is called Regular Grid or Ordered Grid. It can be pictured as a grid that has higher resolution than the pixel grid. Where SSAA samples in each square in that grid, SGSSAA samples in only some of the squares.
SSAA is slow because the pixel rendering stage of the rendering pipeline must do the same amount of work as if the screen resolution was many times the resolution of what it is.
 

boxleitnerb

Platinum Member
Nov 1, 2011
2,605
6
81
So you imply that AMDs SSAA is an ordered grid SSAA? Can you prove that with FSAA Viewer screenshots?

I'm saying "SSAA" doesn't exist because if you mean ordered grid, you need to call it that, namely OGSSAA. Maybe we should get that straight before discussing this any further.

From THG:
http://www.google.de/imgres?um=1&hl...58&start=48&ndsp=59&ved=1t:429,r:84,s:0,i:341

This is AMDs 4xSSAA on a HD5870. It is in no way OGSSAA:
EKsqf.png


Looks very similar to the SSAA on my 580 from earlier, just the rotation angle is different:
9CNOl.jpg
 
Last edited:

BFG10K

Lifer
Aug 14, 2000
22,709
3,002
126
AMD’s and nVidia’s RGSS/SGSS are functionally equivalent. The methods to obtain their samples may vary slightly (as does their positioning), but the fundamentals are exactly the same. That’s why it’s called RGSS/SGSS.

Furthermore, RGSS/SGSS generally offers superior image quality to OGSS at identical sample counts. It’s OGSS that’s the brute force method, so it doesn’t look as good, and also runs slower.

Also when people talk about sampling “some of the squares” or something, I don’t think they really understand what’s happening.
 
Last edited:

boxleitnerb

Platinum Member
Nov 1, 2011
2,605
6
81
Thanks for the explanation.
What is the difference between RGSS and SGSS - if there is any?

I had a friend with a HD7950 make a screenshot with 4xSSAA:


And the 580 for comparison.
9CNOl.jpg


It seems HD7000 SSAA is different from HD5000 SSAA. And it looks exactly the same as Nvidias SSAA.
 
Last edited: