4-way SLI 680 and BF3 MP = ~70% GPU Usage

S

sk3tch

2[H]4U
Joined
Sep 5, 2008
Messages
3,393
08/23/12 UPDATE - looks like it is a driver issue with NV Surround that creates a GPU bound situation. It is not CPU bound. Reference this post.

ORIGINAL POST
So with 4-way SLI 680s (131% power, +0% core, +450% mem, max software voltage) and a 3960X CPU (4.5 GHz), plus I'm running at PCIe 3.0 - still seeing (on average) 70% usage per GPU and frames that vary from 50 FPS up to 90+ FPS in BF3 multiplayer at 5760x1080 (NV Surround). EDIT: I also have "prefer maximum performance" set in the NVIDIA control panel.

Has anyone else experienced this? I can't believe this would be a CPU bottleneck issue. I guess it must be drivers? I'm on 304.48 and I don't want to go up to 304.79 as I had issues with it (pretty bad GPU scaling).

Before anyone asks, here's a quick, crappy pic:

(click to enlarge)
/ORIGINAL POST

08/23/12 UPDATE - looks like it is a driver issue with NV Surround that creates a GPU bound situation. It is not CPU bound. Reference this post.
 
Last edited:
I've seen a video where this is the case. It's kinda disappointing, I know. Maybe they are behind in improving NV surround performance in their drivers, like you said.
 
4 way SLI scaling with the 600 series is not where it needs to be yet. Even 3 way is scaling horribly with 680's in the current drivers. I'm kinda surprised because 3 way scaling with 580s is pretty much perfect. Must be because it's a whole new architecture and the drivers are immature. Give it some time and Nvidia will get it right. In the mean time, If I were you, I would remove one or two of your cards and try SLI and 3 way and see what kind of results you get. I thought I remember reading a post where someone said they had higher FPS in BF3 with 3 680s after removing the fourth. Worth a shot
 
willdearborn said:
4 way SLI scaling with the 600 series is not where it needs to be yet. Even 3 way is scaling horribly with 680's in the current drivers. I'm kinda surprised because 3 way scaling with 580s is pretty much perfect. Must be because it's a whole new architecture and the drivers are immature. Give it some time and Nvidia will get it right. In the mean time, If I were you, I would remove one or two of your cards and try SLI and 3 way and see what kind of results you get. I thought I remember reading a post where someone said they had higher FPS in BF3 with 3 680s after removing the fourth. Worth a shot
Click to expand...

Thanks for the reply. Yeah, that may have been me, actually. I previously had 4x FTW+ 4GB cards (sold/selling them due to bad thermal performance in 4-way - reference was much better) and I had to RMA one so I was running 3-way for a while. The performance was nearly on par with 4-way in BF3 MP. You maybe gain 10% from adding that 4th card in BF3 MP.

By the time 680 4-way is optimized the new gen will be out - and I'll probably be upgrading to those. Heh. Just wondering if I have something set wrong or something so I'm donig a gut check.
 
Anything beyond 2 680's is seeing very little performance boost.... 25%-35% for 3rd, another 15-20% for a 4th.

Not worth an extra $1000

Its drivers. Well known right now.
 
adam_the_giant said:
Anything beyond 2 680's is seeing very little performance boost.... 25%-35% for 3rd, another 15-20% for a 4th.

Not worth an extra $1000

Its drivers. Well known right now.
Click to expand...

Word. I get about 80% of the performance of my 4-way SLI with my 2-way SLI. Just making sure I'm not missing something here.
 
sk3tch said:
Word. I get about 80% of the performance of my 4-way SLI with my 2-way SLI. Just making sure I'm not missing something here.
Click to expand...

Incredible amount of power though.

I just sold my 2nd 7970 as I game at 1080p and 2x7970's was overkill for my needs.
 
adam_the_giant said:
Incredible amount of power though.

I just sold my 2nd 7970 as I game at 1080p and 2x7970's was overkill for my needs.
Click to expand...

Yeah, plus I was not happy with how my 7970s performed in CFX. AMD is making gains in drivers, but NVIDIA is still where it's at.
 
I had completely normal utilization with 3 classys... so I'll see what happens when I drop in the 4th.
 
I thought it was well-established that 3-way xfire and SLI didn't really add much in framerate but really reduced microstuttering?
 
That's_Corporate said:
while i've HEARD the same, i have yet to see proof (in terms of reducing microstuttering)
Click to expand...

True. Also - microstuttering often comes down to each users sensitivity aka their perception of the stutter (along with other factors such as monitor refresh rate to FPS, etc).

That said, performance definitely increases as you add GPUs. There's just less effective scaling beyond 2 cards.
 
fomoz said:
It's a driver issue. 4-way SLI Surround doesn't work well right now.
Click to expand...

I don't buy into your specific theory, but I do think it's an issue with driver optimization for the game. This isn't Unigine Heaven or 3DMark 11...it's Battlefield 3 multiplayer. And the only way to push this many pixels is via NV Surround or Eyefinity so there's not a way to compare as easily.
 
This is just standard with 4 way SLI unfortunately..you will have LOTS of scaling and usage issues ......4 way is great for benchmarks but a lot of games have issues with scaling beyond 2 cards.

What kind of OCs do you get on your classys btw?
 
So far I have +50 on the core and +450 mem across all cards working flawlessly during BF3 MP. Pretty happy with it. Best 4-way experience yet for me.
 
sk3tch said:
True. Also - microstuttering often comes down to each users sensitivity aka their perception of the stutter (along with other factors such as monitor refresh rate to FPS, etc).

That said, performance definitely increases as you add GPUs. There's just less effective scaling beyond 2 cards.
Click to expand...

i did a bit of research, and this is what i found; it basically comes down to being CPU bound.

the reason tri-SLI/tri-fire reduces microstuttering is because the cards work less and the CPU becomes the limiting factor. if you think about it, two cards may work at 90% efficiency/scaling while three cards will see that drop to about 70%. with less work being done, the limit of the CPU becomes apparent and stuttering is reduced greatly.

this is also why frame limiting essentially removes microstuttering, it's just done artificially, instead of physically by adding another card. setting the FPS limit to 59 greatly reduces the load on the cards, increasing the probability of being CPU bound.


if any of that was way off base, i apologize! but it makes sense to me :)
 
What utter sillyness is spread here...

It is NOT the drivers. Of course it is a CPU bottleneck in BF3 MP. And if you play with MSAA/FXAA, in most other titles as well.

People...for the love of god, I cannot read this anymore. Test it yourself before coming here to complain. Anyone who has the money to buy more than 2 cards is expected to know how to find a CPU bottleneck.
 
That's_Corporate said:
i did a bit of research, and this is what i found; it basically comes down to being CPU bound.

the reason tri-SLI/tri-fire reduces microstuttering is because the cards work less and the CPU becomes the limiting factor. if you think about it, two cards may work at 90% efficiency/scaling while three cards will see that drop to about 70%. with less work being done, the limit of the CPU becomes apparent and stuttering is reduced greatly.

this is also why frame limiting essentially removes microstuttering, it's just done artificially, instead of physically by adding another card. setting the FPS limit to 59 greatly reduces the load on the cards, increasing the probability of being CPU bound.


if any of that was way off base, i apologize! but it makes sense to me :)
Click to expand...

Thanks!

boxleitnerb said:
What utter sillyness is spread here...

It is NOT the drivers. Of course it is a CPU bottleneck in BF3 MP. And if you play with MSAA/FXAA, in most other titles as well.

People...for the love of god, I cannot read this anymore. Test it yourself before coming here to complain. Anyone who has the money to buy more than 2 cards is expected to know how to find a CPU bottleneck.
Click to expand...

Thanks for the insight. This is why I came here - for advice/help. I did not realize that I was expected to know things just because I own a certain amount of video cards. :cool:

Vega said:
http://www.youtube.com/watch?v=tkZzssm-kWs

http://www.youtube.com/watch?v=IyQoFPO8RyA

Both my 4-way 680 and 7970 setups reach 90+% utilization routinely. Even on full 64-player servers I manage 120+ FPS at all times. Then again, my CPU's are always around 5 GHz.
Click to expand...

Great to know, thank you! I will push my CPU higher. I can get it to 4.9 GHz but it requires 1.5v.


fomoz said:
BF3: How to tell whether your CPU or GPU is Lagging You
Click to expand...

Thank you for this. I did try this at one point - but I was not sure what to look for. To me, it looks like I was not having CPU issues. I figured at 4.5 GHz and 6 cores (SB-E) that I would be fine. Off I go for more tests...I just was hoping to not have to push so many volts.
 
Sorry for the rough tone - it's just that this issue comes up so often and the answer is always the same. Test for a CPU bottleneck. If it's not that, then we can move onto a driver problem.

CPU performance has increased slower than GPU performance the last decade. Now if you put 4 cards in, the imbalance will be even worse. The use of 3-way and 4-way SLI is gameplay with the fastest OCed CPUs one can have or with advanced image quality settings like 3D Vision, SSAA, downscaling, SSAO etc. You don't have to overclock your CPU so massively to utilize your cards. If you use the mentioned IQ settings, the problem takes care of itself.

5760x1080 is nice, but it is only 3x the amount of pixels than 1080p. Performance doesn't scale linearly, meaning at 5760x1080, performance will not be 1/3 but a little bit more. Add to that you have 4 cards, not 3, so you have extra GPU power that cannot be brought to bear.
 
No problem. I know how it is. People on forums our generally not the brightest bunch - present company excluded, of course! ;)

I just played BF3 MP for about 15 minutes with my 3960X cranked to 4.9 GHz/1.5v (previously: 4.5 GHz/1.45v) - approximately the same results. :( Good news is my CPU temps were good...one core maxed out at 73C otherwise the rest were under 70C.
 
Last edited:
10% is difficult to reproduce in multiplayer. I would just live with it and be happy to have some reserves for other more demanding games.
 
boxleitnerb said:
What utter sillyness is spread here...

It is NOT the drivers. Of course it is a CPU bottleneck in BF3 MP. And if you play with MSAA/FXAA, in most other titles as well.

People...for the love of god, I cannot read this anymore. Test it yourself before coming here to complain. Anyone who has the money to buy more than 2 cards is expected to know how to find a CPU bottleneck.
Click to expand...

This is of course a bunch of nonsense and i'm not sure why you're so upset at the suggestion. Quad and Tri SLI doesn't scale properly in many titles and this is an absolute fact, whereas 2 way sli works in nearly every title. There are benchmarks on various websites proving this, showing negative scaling in a mind boggling amount of games, I could dig it up if you wish - and yes I have fiddled with tri in the past. The question is, are you using tri sli? I have a feeling the answer is no because if you have, you'd find these issues yourself. These issues are why I will never go past 2 way.

BF3 does scale well with 2 GPUs and you are probably right that it is a CPU issue in OPs case. But many other titles have issues in tri or quad sli.
 
boxleitnerb said:
10% is difficult to reproduce in multiplayer. I would just live with it and be happy to have some reserves for other more demanding games.
Click to expand...

Maybe you just have too much power for the resolution you are using lol.

Maybe go with 3x 1600p monitors?
 
xoleras said:
This is of course a bunch of nonsense and i'm not sure why you're so upset at the suggestion. Quad and Tri SLI doesn't scale properly in many titles and this is an absolute fact, whereas 2 way sli works in nearly every title. There are benchmarks on various websites proving this, showing negative scaling in a mind boggling amount of games, I could dig it up if you wish - and yes I have fiddled with tri in the past. The question is, are you using tri sli? I have a feeling the answer is no because if you have, you'd find these issues yourself. These issues are why I will never go past 2 way.

BF3 does scale well with 2 GPUs and you are probably right that it is a CPU issue in OPs case. But many other titles have issues in tri or quad sli.
Click to expand...

I'm upset because the most likely explanation that is easiest to test for is almost always overlooked. "No, my CPU cost $1000, that cannot be it..." Read my explanation again. Test the game in the lowest HD-resolution - that is the maximum amount of fps your CPU will provide in that scene, no matter what resolution, AA and number of GPUs you throw at the game. With that you can work.

While there surely are titles where scaling isn't perfect, a CPU bottleneck is a far more likely explanation. Be my guest, dig out your evidence. If the cards are properly taxed, they mostly scale well. You have to understand that benchmarks are often a mix of GPU and CPU bottlenecks. If you are not completely GPU bound, scaling will of course suffer.

I am using 2-way SLI but with GK110 I will be going 3-way for sure. Also, IF scaling is bad in a title, one can always experiment with the SLI compatibility bits. I know some cases where this has helped tremendously.
 
Last edited:
xoleras said:
BF3 does scale well with 2 GPUs and you are probably right that it is a CPU issue in OPs case. But many other titles have issues in tri or quad sli.
Click to expand...

Well, I moved my 3960X from 4.5 GHz to 4.9 GHz and re-tested - basically the same results. 400 MHz more juice x 6 cores you would think would change things somewhat? At this point my "guess" is drivers. I don't know what bottleneck I could have...fast RAM, fast SSD RAID 0, fast CPU, fast GPU...motherboard is the ASUS Rampage IV Extreme (doubt that matters) and I monitor GPU usage and FPS while in-game along with monitoring temps. Not seeing any thermal throttling at 70C as I max out at 66C with my GPUs.


DASHlT said:
Maybe you just have too much power for the resolution you are using lol.

Maybe go with 3x 1600p monitors?
Click to expand...

My resolution is "low" but they're 120hz monitors so my goal is 90 FPS+ for online multiplayer fluidity.
 
How do you test in MP? Are you sure your results are repeatable? 400MHz is less than 10% more - if the results are not completely repeatable that may be "overlooked" quite easily.

My suggestion (with the 4.9GHz setting and multiplayer):

  1. Have EVGA Precision or MSI Afterburner OSD running, showing CPU clock (disable turbo) and GPU usage.
  2. Go to a spot where your fps are as stable as possible.
  3. Alt-Tab out of the game and go into windows energy options. There lower the "maximum processor state":

    This will change your CPU clock "live" without having to reboot and changing the multi in the bios.
    Check the CPU clock with CPU-Z. If the CPU is idle because you're out of the game, use Prime/Linx etc.
  4. Go back into the game and make screenshots (same spot, don't move, don't move the mouse) with different CPU clocks, for example 4.9, 4.5, 4.0, 3.5 GHz). You might have to play with the percentage a bit, but a good granularity can be achieved.
  5. Post the screenshots here. Then we can see if and how the GPU usage and fps rely on CPU clock.
 
sk3tch said:
Well, I moved my 3960X from 4.5 GHz to 4.9 GHz and re-tested - basically the same results. 400 MHz more juice x 6 cores you would think would change things somewhat? At this point my "guess" is drivers. I don't know what bottleneck I could have...fast RAM, fast SSD RAID 0, fast CPU, fast GPU...motherboard is the ASUS Rampage IV Extreme (doubt that matters) and I monitor GPU usage and FPS while in-game along with monitoring temps. Not seeing any thermal throttling at 70C as I max out at 66C with my GPUs.




My resolution is "low" but they're 120hz monitors so my goal is 90 FPS+ for online multiplayer fluidity.
Click to expand...

Are you getting a constant 120fps in bf3 when you play?
 
boxleitnerb said:
How do you test in MP? Are you sure your results are repeatable? 400MHz is less than 10% more - if the results are not completely repeatable that may be "overlooked" quite easily.

My suggestion (with the 4.9GHz setting and multiplayer):

  1. Have EVGA Precision or MSI Afterburner OSD running, showing CPU clock (disable turbo) and GPU usage.
  2. Go to a spot where your fps are as stable as possible.
  3. Alt-Tab out of the game and go into windows energy options. There lower the "maximum processor state":

    This will change your CPU clock "live" without having to reboot and changing the multi in the bios.
    Check the CPU clock with CPU-Z. If the CPU is idle because you're out of the game, use Prime/Linx etc.
  4. Go back into the game and make screenshots (same spot, don't move, don't move the mouse) with different CPU clocks, for example 4.9, 4.5, 4.0, 3.5 GHz). You might have to play with the percentage a bit, but a good granularity can be achieved.
  5. Post the screenshots here. Then we can see if and how the GPU usage and fps rely on CPU clock.
Click to expand...

Thank you very much. I will try this as soon as I am able (at work).


DASHlT said:
Are you getting a constant 120fps in bf3 when you play?
Click to expand...

No sir. Per my original post - it bounces from 50 FPS to 90+ FPS depending on the online multiplayer conditions. My goal is as close as I can get to 90 FPS+ constant at 5760x1080.
 
sk3tch said:
So with 4-way SLI 680s (131% power, +0% core, +450% mem, max software voltage) and a 3960X CPU (4.5 GHz), plus I'm running at PCIe 3.0 - still seeing (on average) 70% usage per GPU and frames that vary from 50 FPS up to 90+ FPS in BF3 multiplayer at 5760x1080 (NV Surround). EDIT: I also have "prefer maximum performance" set in the NVIDIA control panel.

Has anyone else experienced this? I can't believe this would be a CPU bottleneck issue. I guess it must be drivers? I'm on 304.48 and I don't want to go up to 304.79 as I had issues with it (pretty bad GPU scaling).

Before anyone asks, here's a quick, crappy pic:

(click to enlarge)
Click to expand...

4way 7970 achives 200fps so I assume either drivers or some conflict with hardware somewhere.
 
Flopper said:
4way 7970 achives 200fps so I assume either drivers or some conflict with hardware somewhere.
Click to expand...

Yea if you arent getting a full 120fps at the res with the 4 card something is wrong.

Try out 3 cards and see if you get 120fps
 
Flopper said:
4way 7970 achives 200fps so I assume either drivers or some conflict with hardware somewhere.
Click to expand...

I think you may not be factoring resolution into this. At 1080p, I'm sure I can get 200 FPS. I'm running at 5760x1080 at 120hz.


DASHlT said:
Yea if you arent getting a full 120fps at the res with the 4 card something is wrong.

Try out 3 cards and see if you get 120fps
Click to expand...

Thanks...but I have used 3-way SLI (with my prior FTW+ 680 4GB cards - sold them due to poor thermal performance) and the performance between 3-way and 4-way in BF3 MP is pretty similar. You only gain about 10% by adding that last card (all based on eyeballing FRAPS and no scientific results). Which is why I'm inclined to think it's poor optimization. But I am definitely going to look at a CPU bottleneck.


Homer said:
And try three cards with .79 drivers, maybe...
Click to expand...

Thanks. I'm hesitant to try .79 (tried it with my last 4-way setup a few weeks ago and I had issues with mGPU) but I definitely will after I do the tests outlined by boxleitnerb.
 
Last edited:
I had issues with .79 and quad sli 690's, so I just started with .48 on my classys to save myself driver headaches. Just curious to see though, because I know they've fixed some BF3 performance issues in more recent drivers.
 
Something doesn't sound right with the setup to have that low of FPS numbers. I get 120-190 FPS pushing 5x 120Hz 1080P screens with four 7970's in BF3 multi-player. Granted, all settings aren't maxed. Are you trying to max all settings, even silly 4x MSAA?

Oh BTW turn mesh quality to low, that is a huge CPU hog. I bet your GPU usage and FPS get a decent boost by doing that.
 
You must log in or register to reply here.
Back
Top