Retested with the GTX 660:
OS: Win 8 Professional x64
CPU: Intel Core i5 3570K (stock, 3.7 GHz with turbo boost)
GPU: Nvidia GTX 660
Ram: 2 x 4GB (8GB) DDR3 1600 MHz
Dolphin build: 64 bit
Game region: NTSC
Scene 1: 231-251%
Scene 2: 191-203%
Scene 3: 276-316%
As you can see the delta increases as the speed gets higher. That's because at 300% speed a 1% difference in speed changes the counter by 3%.
GPU testing:
I did some additional testing at different resolutions as well. I'm not going to bother clogging up the thread with a zillion images so I'm just including the 1x images above.
Number inside parenthesis indicates the actual IR relative to native. % are the average. They are not calculated in any accurate way but are simply my estimations based on observation.
1x (1x):
241%
198%
296%
2x (4x):
241%
198%
296%
2x shows no difference compared to 1x.
3x (9x):
235%
198%
296%
3x shows no difference compared to 2x.
2x + 4xSSAA (16x):
235%
198%
296%
2x + 4xSSAA shows no difference from 3x.
3x + 4xSSAA (36x):
195%
195%
195%
Very odd results but they are accurate. All three now run at 195% speed average with almost no wavering up/down. Scenes 1 and 3 took a fairly large plummet while scene 2 stayed the same. Do keep in mind though that the game is still running at twice the native speed.
3x + 9xSSAA (81x):
88%
130%
88%
Big plummet in scenes 1 and 3, moderate plummet in scene 2. We're now below fullspeed. This confirms that scenes 1 and 3 are much more demanding on the GPU than scene 2.
And just for shits and giggles
3x + 9xSSAA + per-pixel lighting:
88%
130%
88%
Ok this is weird. Considering how high my internal resolution is and how few polygons are in these three scenes per-pixel lighting should be WAY slower than per-vertex lighting according to common sense. So either it's not working or my understanding of the option is flawed. Perhaps it is because dolphins video backends are limited entirely by video memory bandwidth and not shader throughput on modern GPUs as I have always suspected.
I'll post openGL and d3d11 results soon.
Edit: The results are in.
D3d11 1x:
242%
198%
300%
D3d11 2x:
242%
198%
300%
D3d11 3x:
242%
198%
300%
OpenGL 1x:
241%
197%
298%
OpenGL 2x:
241%
197%
298%
OpenGL 3x:
241%
197%
298%
They are all exactly the same. 1x, 2x, or 3x. D3d9, d3d11, or openGL. All yield the exact same results +/- 1 or 2%. This is because the GPU is not being bottlenecked with any of them. I need to test with 5x or higher to truly see how the three backends perform. I could stress it with MSAA both those results would not be comparable to SSAA.
Edit 2: The results with max MSAA/CSAA are exactly the same in d3d11 and openGL. It still isn't pushing the card hard enough.
OS: Win 8 Professional x64
CPU: Intel Core i5 3570K (stock, 3.7 GHz with turbo boost)
GPU: Nvidia GTX 660
Ram: 2 x 4GB (8GB) DDR3 1600 MHz
Dolphin build: 64 bit
Game region: NTSC
Scene 1: 231-251%
Scene 2: 191-203%
Scene 3: 276-316%
As you can see the delta increases as the speed gets higher. That's because at 300% speed a 1% difference in speed changes the counter by 3%.
GPU testing:
I did some additional testing at different resolutions as well. I'm not going to bother clogging up the thread with a zillion images so I'm just including the 1x images above.
Number inside parenthesis indicates the actual IR relative to native. % are the average. They are not calculated in any accurate way but are simply my estimations based on observation.
1x (1x):
241%
198%
296%
2x (4x):
241%
198%
296%
2x shows no difference compared to 1x.
3x (9x):
235%
198%
296%
3x shows no difference compared to 2x.
2x + 4xSSAA (16x):
235%
198%
296%
2x + 4xSSAA shows no difference from 3x.
3x + 4xSSAA (36x):
195%
195%
195%
Very odd results but they are accurate. All three now run at 195% speed average with almost no wavering up/down. Scenes 1 and 3 took a fairly large plummet while scene 2 stayed the same. Do keep in mind though that the game is still running at twice the native speed.
3x + 9xSSAA (81x):
88%
130%
88%
Big plummet in scenes 1 and 3, moderate plummet in scene 2. We're now below fullspeed. This confirms that scenes 1 and 3 are much more demanding on the GPU than scene 2.
And just for shits and giggles
3x + 9xSSAA + per-pixel lighting:
88%
130%
88%
Ok this is weird. Considering how high my internal resolution is and how few polygons are in these three scenes per-pixel lighting should be WAY slower than per-vertex lighting according to common sense. So either it's not working or my understanding of the option is flawed. Perhaps it is because dolphins video backends are limited entirely by video memory bandwidth and not shader throughput on modern GPUs as I have always suspected.
I'll post openGL and d3d11 results soon.
Edit: The results are in.
D3d11 1x:
242%
198%
300%
D3d11 2x:
242%
198%
300%
D3d11 3x:
242%
198%
300%
OpenGL 1x:
241%
197%
298%
OpenGL 2x:
241%
197%
298%
OpenGL 3x:
241%
197%
298%
They are all exactly the same. 1x, 2x, or 3x. D3d9, d3d11, or openGL. All yield the exact same results +/- 1 or 2%. This is because the GPU is not being bottlenecked with any of them. I need to test with 5x or higher to truly see how the three backends perform. I could stress it with MSAA both those results would not be comparable to SSAA.
Edit 2: The results with max MSAA/CSAA are exactly the same in d3d11 and openGL. It still isn't pushing the card hard enough.
"Normally if given a choice between doing something and nothing, I’d choose to do nothing. But I would do something if it helps someone else do nothing. I’d work all night if it meant nothing got done."
-Ron Swanson
"I shall be a good politician, even if it kills me. Or if it kills anyone else for that matter. "
-Mark Antony
-Ron Swanson
"I shall be a good politician, even if it kills me. Or if it kills anyone else for that matter. "
-Mark Antony