GTX 680 x3 - Extremely Poor BF4 Performance

[UaVaj]

already thought of that too. however in game performance and feel is exact same. rather that be 1 gpu or 2 gpu or 3 gpu. same stuttering mess due to low fps (~20 fps). especially apparent at 5760x1080.

if you watch my videos (link on post #1). you will see that all gpu is well loaded.

something is obviously broken.

-----

want to separate the sheep from the goats? disable sli and report back. does it will feel the exact same?

at 1600x1000 you are beyond 60fps - rather that is with 1 gpu or 2 gpu.

hopefully someone with a 2560x1440p monitor and with SLI (x2 or x3 or x4) will further clarify.

 

[mindbomb]

I believe your issue is that the extremely high resolution prevents sli connection from working properly. Basically the reason there is no crossfire connector on the r9 290.

 

[UaVaj]

please pay attention to post #1. it is very through and with video and constantly being updated.

this issue happens at both 5760x1080 and 1920x1080. hopefully i can get my hands on a 2560x1440 and will add that data too.

 

[2is]

already thought of that too. however in game performance and feel is exact same. rather that be 1 gpu or 2 gpu or 3 gpu. same stuttering mess due to low fps (~20 fps). especially apparent at 5760x1080.

if you watch my videos (link on post #1). you will see that all gpu is well loaded.

something is obviously broken.

-----

want to separate the sheep from the goats? disable sli and report back. does it will feel the exact same?

at 1600x1000 you are beyond 60fps - rather that is with 1 gpu or 2 gpu.

hopefully someone with a 2560x1440p monitor and with SLI (x2 or x3 or x4) will further clarify.

Here it is with SLI disabled:

It felt largely the same, with the exception of when I'm in the middle of a large explosion, I could feel a performance drop and a quick glace at my FPS showed it in the 50's during those explosions.

 

[UaVaj]

thanks for that single gpu screenshot. definitely de ja vu.

it is clearly time to let NVidia and Dice figure this out.

wonder if AMD has this same scaling issue? any crossfire folks willing to post screenshots with "perfoverlay.drawgraph 1" for comparison?

 

[2is]

thanks for that single gpu screenshot. definitely de ja vu.

it is clearly time to let NVidia and Dice figure this out.

wonder if AMD has this same scaling issue? any crossfire folks willing to post screenshots with "perfoverlay.drawgraph 1" for comparison?

Well, lets look at this from another angle.... Maybe I misunderstood your explanation about the frame times, but based on what I understood, that would mean I'm actually getting higher performance with a single card, which I know not to be the case. Similarity large explosions where I dropped to the 50's on a single card would still be in the 80's with SLI enabled.

From my understanding, latency is always increased with SLI, that's just the nature of the beast. The key is to have a performance increase that can offset that added latency (and then some) which based on my observations, is exactly what is happening.

 

[UaVaj]

your two screenshots are at different map and different load. even so, based on the "graph's gpu frame times." you are indeed getting slightly higher performance with a single 680. hence the de ja vu.

given that you are gaming at 1600x1000. suspect that those frame dips during explosions are due to cpu bottleneck. not gpu bottleneck. that is why I suggested someone with 2560x1440 to add more data.

as for your frame dip. are you measuring that dip with the counter on the upper right? if so. that counter is a cpu frame counter. not gpu.

better way is to record the video with shadow play and pause it at the moment of the explosion and compare both the dip in cpu and gpu from the graph. that will be objective.

 

[jackstar7]

Running Tri-Fire 7990+7970 I'm seeing average GPU of 12.5ms in action and down to as low as 7ms in non-action. Max was showing about 15-20ms depending on the time I was looking.

This is with Win7 and Mantle and all 3 GPUs active. If I have time to get into the game again, I'll try and mess around and get screens. This was just jumping in for a moment to check against your findings.

Also, I run with a Strobelight-ed 1080p, so I do see the difference above 60fps.

 

[2is]

I'm at 1920x1200... Those same explosions don't dip anywhere near as low when SLI is enabled, using the same CPU.

As stated, multi GPU always adds latency. If your interpretation of frame times is accurate, that means 1 card will always be faster than two, which we know isn't the case.

Based on that, I have to ask... Are you actually getting poor performance? Or looking at frame times and thinking you're getting poor performance?

To me everything is working the way it should...

SLI enabled = latency increased = normal (it would be nice if it wasn't, but it is what it is with todays technology)
SLI enabled = higher performance = observed in actual gameplay, verified by fraps
SLI enabled = higher CPU load by way not not being bottlenecked by single GPU = observed
SLI enabled = observed both GPU's being loaded

I'm not seeing an issue here.

 

[UaVaj]

jackstar7
looking forward to those screenshots.
12.5 ms is proof multi gpu is working. that is 80fps solid.
7.5 ms is 133fps.



2is
your screenshots was resizing to 1600x1000.

anyway at 1920x1080. a single 680 has more grunt that my eyes can detect. anything above 45fps and I am good.
likewise on the kids machine - i5-760 oc 3.8GHz w/ a 680 2GB is hovering at 70 fps and dipping to 55 fps with max ultra setting. that is fluid.

my machine at 5760x1080. 680 x1, 680 x2, 680 x3 all have the same stuttering mess. all gpu frame times showing ~20fps. clearly not going by frame time only. it is unplayable.

again. fraps is measuring cpu frame time. not gpu frame time.
those screenshots you posted are undisputable.

the last nail on this coffin is to compare jackstar7 screenshots.

 

[Majcric]

Anybody googled to see what the consensus on Sli is in general with BF4? Of course, there will only be a few playing at the OP rez.

 

[mindbomb]

So when you checked 1080p, that was with a 100% resolution scale, right?

 

[Majcric]

Op, are you using 100% max settings.

I'm thinking you could almost get by on a heavily overclocked 780ti. You're at 3k and said you couldn't see above 45fps

https://www.youtube.com/watch?v=vO8C7IgKEjs

 

[UaVaj]

mindbomb
yes. both 5760x1080 and 1920x1080 are tested with 100% resolution scale (no ssaa)

Majcric
yep. go big or go home. only play at max settings (aka ultra settings).
at 5760x1080. a single 780ti classified "should" be able to deliver playable frame rate with some dipping at high setting. as for ultra setting - no way.

 

[Majcric]

it's pretty much 680sli, give it a go. Did you watch the video?

 

[Hitman928]

I'm guessing your experience has correlation to pcper's results in the 295x review?

http://www.pcper.com/reviews/Graphi...-295X2-8GB-Graphics-Card-Review/Battlefield-4

 

[UaVaj]

hitman928.

thanks for the link.

780ti x2 is a stuttering mess. frame variance is off the chart. reminds me of 7970 cross broke.

looks like if I want to continue playing bf4. nvidia is out the window this round.

unless there is already a fix in the pipeline. NVidia can kiss goodbye my $3k gpu budget on three/four 780ti 6gb.

 

[moonbogg]

For what its worth, my two 670's seem to work well with BF4. There are occasional anomalies, where SLI straight up disengages and stops working. Yes, this actually happens once in a while and I've noticed it mostly on Hainan resort, so that might not be your issue. I have noticed a stuttering feeling with SLI and this game though, despite high FPS in the 70-80 fps range. The game just doesn't always feel smooth like it should, but I gave up and just play. Screw it.

 

[UaVaj]

part of this is perception. can you perceive the difference? even when SLI is broken.

at 1920x1080. cause a single 680 is delivering more fps than my eyes can detect. I cannot tell difference between 680 x1, 680 x2, 680 x3. regardless if SLI is working or not. otoh - the cpu/gpu graph clearly shows the truth. the truth in which I was unable to detect. that SLI is in fact broken.

at 5760x1080. due to the low fps from SLI being broken. I am able to easily detect the frame drop. SLI is obviously broken.

-------

unless someone what to prove me wrong with some screenshots and/or videos.

 

[n0x1ous]

I get great SLI performance with my 670's. I play @ 120hz and normally i am between 90-120hz on a full map. gpu usage is very even across the cards- 95% on both when getting up to 120fps and usually in the 80% when dropping down to 90 FPS from what i assume is a CPU bottleneck... I only play conquest large 64 player......

 

[UaVaj]

please do post screenshots or video
with both perfoverlay.drawgraph 1 and perfoverlay.drawfps 1

 

[n0x1ous]

please do post screenshots or video
with both perfoverlay.drawgraph 1 and perfoverlay.drawfps 1

I'll try to grab some proof tonight after work.

And to be clear. settings are 1080p@120hz. VG248QE with 2d lightboost. All settings Ultra + 4xMSAA - FXAA is off - Resolution scale 100%

337.50 beta drivers. Windows 8.1 x64 (though I had no issues on previous drivers either).

 

[UaVaj]

just for the record. not doubting you at all. just need more "objective" data to for my cause.

have a 3k budget for three/four 780ti 6gb that NVidia may not get a penny depending on how broken SLI is for BF4.

right now AMD is looking real real good.

looking forward to those screenshots.

 

[n0x1ous]

For what its worth, my two 670's seem to work well with BF4. There are occasional anomalies, where SLI straight up disengages and stops working. Yes, this actually happens once in a while and I've noticed it mostly on Hainan resort, so that might not be your issue. I have noticed a stuttering feeling with SLI and this game though, despite high FPS in the 70-80 fps range. The game just doesn't always feel smooth like it should, but I gave up and just play. Screw it.

Moon - we have similar gpu setups and I am getting a little better numbers than you.....you have a better cpu but clocked a little slower. my 670's are usually boosting between 1110 and 1163 (top/bottom) card once they are warmed up.....curious about your clocks...

 

[n0x1ous]

just for the record. not doubting you at all. just need more "objective" data to for my cause.

have a 3k budget for three/four 780ti 6gb that NVidia may not get a penny depending on how broken SLI is for BF4.

right now AMD is looking real real good.

looking forward to those screenshots.

yeah no prob. totally understand. I would want to see evidence too if I was in your shoes. Curious if you have tried it with just 2 cards vs 3?