- Dec 2, 2011
- 397
- 24
- 81
I finally bought another 4GB Asus GTX670 card and set it up in SLI. I inserted it in the second PCI-e slot in my P8P67 Deluxe, connected power cables and connected the SLI bridge.
Booted up and all looks good. Enabled SLI in drivers (314.22). Note! I did a fresh Windows-install after bying and inserting the first 670. I did not re-install drivers after inserting the second card.
But when I started testing things were not so good. It seems there is no scaling at all, or just slightly.
Anyway, here are the test results (sorry for slow ImageShack). Unless specified, all test were run at 1920x1080 with AA off and AF @ 16x
Just Cause 2:
This is single-GPU with AA 4x. All settings are max, but SSAO, HDshadows and Water emu are off. The usage is almost at 100% and the fps 132.
This is SLI (Alternate Frame Rendering 2) with AA 4x. Almost identical performance here. No wonder, since both GPUs are used only 50%. 0.5 + 0.5 = 1 isn't it? I used AFR2 here, because the "nVidia recommended" setting actually produced even lower fps in SLI. AFR1 was abysmal.
Far Cry 3:
Settings are mostly at medium or high, geometry at ultra. I only have an image of SLI here, but single-GPU yields almost identical performance in this spot. At the base of the nearby antenna, when overlooking the village, I get 83 in single and 93 in SLI. So a slight increase in that one spot, but far from what I expected.
Skyrim: Yeah I know this game sometimes has negative SLI scaling, but I'm giving an image here anyway, since it follows the trend from other games. Even in one of the most demanding spots (on top of stairs at Dragonsreach) it has exacly the same fps with or without SLI. Image is of SLI. At most demanding spot in Markarth I also get about the same fps with or without SLI, around 46-50.
Crysis benchmark:
I set all settings to Gamer and ran the standard GPU benchmark of the vanilla game. Here are the results:
SLI:
Play Time: 21.79s, Average FPS: 91.76
Min FPS: 57.58 at frame 197, Max FPS: 120.45 at frame 148
Single-GPU:
Play Time: 22.29s, Average FPS: 89.73
Min FPS: 51.47 at frame 178, Max FPS: 114.76 at frame 1651
In every game I made sure that "Prefer maximum performance" was enabled and of course Single-GPU, nVidia Recommended (SLI) or AFR2, depending on what I was testing.
So does anyone have a clue what is happening here? One should almost thing that my CPU is limiting the cards, but that seems a bit strange since it's running at 4.4GHz (well maybe except Skyrim, I suppose that one is CPU limited). I have seen reviews and youtube videos where they get decent scaling with setup and settings not too far from my own.
Is it a driver issue? Should I reinstall the driver or perhaps try the new 320 driver? Or, given that I'm new to the driver control panel, are there some settings that I have missed?
Need some input here. Sorry for long post.
Edit: I'll have a run-through with Rivatuner or something tonight, just to check that the CPU is in fact running at 4.4GHz.
Edit2: Here is a review of 670 SLI where they even use a 2500 at stock speed it seems. So I guess any CPU limiting theory goes out the Window, at least in this game.
http://hexus.net/tech/reviews/graphics/39605-nvidia-geforce-gtx-670-sli-vs-the-beasttm/?page=6
Booted up and all looks good. Enabled SLI in drivers (314.22). Note! I did a fresh Windows-install after bying and inserting the first 670. I did not re-install drivers after inserting the second card.
But when I started testing things were not so good. It seems there is no scaling at all, or just slightly.
Anyway, here are the test results (sorry for slow ImageShack). Unless specified, all test were run at 1920x1080 with AA off and AF @ 16x
Just Cause 2:
This is single-GPU with AA 4x. All settings are max, but SSAO, HDshadows and Water emu are off. The usage is almost at 100% and the fps 132.
This is SLI (Alternate Frame Rendering 2) with AA 4x. Almost identical performance here. No wonder, since both GPUs are used only 50%. 0.5 + 0.5 = 1 isn't it? I used AFR2 here, because the "nVidia recommended" setting actually produced even lower fps in SLI. AFR1 was abysmal.
Far Cry 3:
Settings are mostly at medium or high, geometry at ultra. I only have an image of SLI here, but single-GPU yields almost identical performance in this spot. At the base of the nearby antenna, when overlooking the village, I get 83 in single and 93 in SLI. So a slight increase in that one spot, but far from what I expected.
Skyrim: Yeah I know this game sometimes has negative SLI scaling, but I'm giving an image here anyway, since it follows the trend from other games. Even in one of the most demanding spots (on top of stairs at Dragonsreach) it has exacly the same fps with or without SLI. Image is of SLI. At most demanding spot in Markarth I also get about the same fps with or without SLI, around 46-50.
Crysis benchmark:
I set all settings to Gamer and ran the standard GPU benchmark of the vanilla game. Here are the results:
SLI:
Play Time: 21.79s, Average FPS: 91.76
Min FPS: 57.58 at frame 197, Max FPS: 120.45 at frame 148
Single-GPU:
Play Time: 22.29s, Average FPS: 89.73
Min FPS: 51.47 at frame 178, Max FPS: 114.76 at frame 1651
In every game I made sure that "Prefer maximum performance" was enabled and of course Single-GPU, nVidia Recommended (SLI) or AFR2, depending on what I was testing.
So does anyone have a clue what is happening here? One should almost thing that my CPU is limiting the cards, but that seems a bit strange since it's running at 4.4GHz (well maybe except Skyrim, I suppose that one is CPU limited). I have seen reviews and youtube videos where they get decent scaling with setup and settings not too far from my own.
Is it a driver issue? Should I reinstall the driver or perhaps try the new 320 driver? Or, given that I'm new to the driver control panel, are there some settings that I have missed?
Need some input here. Sorry for long post.
Edit: I'll have a run-through with Rivatuner or something tonight, just to check that the CPU is in fact running at 4.4GHz.
Edit2: Here is a review of 670 SLI where they even use a 2500 at stock speed it seems. So I guess any CPU limiting theory goes out the Window, at least in this game.
http://hexus.net/tech/reviews/graphics/39605-nvidia-geforce-gtx-670-sli-vs-the-beasttm/?page=6
Last edited: