BF3 CPU bottleneck? Really?

[JAG87]

EDIT: Just did a little testing. I found the most CPU intensive and the most intense part of the entire game I am aware of, sharki on the TV station roof with 64 players. Just standing and overlooking the map.

SLI: 55-60
Single Card: 35-40

SLI: Resolution at 720p, no AA: 60FPS

So I see there is certainly a CPU bottleneck, but even with it the FPS is significantly higher with SLI enabled. CPU usage went up with SLI enabled.

CPU usage with SLI and HT on: 65%
CPU usage with single card and HT on: 50%

FPS while running around sharki with SLI: 80-100
Same thing with single card: 45-60

Are you getting 55-60 at 1080 + MSAA, and 60 at 720 no MSAA?

 

[Termie]

I am impressed. You've found the absolute worst case for your CPU - it will bottleneck at 60fps at that point.

Your averages are so low with a single card - guess it's msaa.

Did you have a chance to check if HT improved minimums or averages? You say CPU use went up, so I bet there was a benefit.

 

[moonbogg]

Are you getting 55-60 at 1080 + MSAA, and 60 at 720 no MSAA?

Correct. FPS doesn't change at that intense location between 1080p totally maxed with 4x MSAA and 720p no AA.

It seems my FPS will never EVER drop far below 60, occasion drops into 50's. No reason to complain, but i'm not done. Below are posted benchmarks from Oman to see SLI benefit with real numbers and to see if minimums are any better with SLI despite the CPU bottleneck. Something special about these runs; they are worst case with chopper blasting rockets right down on me and screen filled with smoke and total chaos with 60-64 players at construction site area. I had the luck of the chopper action being very similar for both runs. This is about the most CPU intense actual gameplay scenario I could find apart from standing in one spot on the hotel or TV station. This reflects CPU limited situations of actual gameplay.

Benches @ 4.6ghz HT on (will do HT off later)

SINGLE CARD
MIN: 38
MAX: 128
AVERAGE: 80

SLI
MIN: 51
MAX: 178
AVERAGE: 103

This is roughly a 25% improvement in a very cpu limited situation. I'll take it. The difference between 38 and 51 is enough to make gameplay much smoother thanks to the additional GPU headroom, but with a faster CPU it could be much better of course.

 

[Termie]

Again, awesome tests. You'll have to wait for a next-gen processor to see better results. You have an amazing setup.

 

[boxleitnerb]

I know the 2600K is fast, but that doesn't mean it isn't bottlenecking. How can I test and measure to actually know what is happening? Should I try to OC my CPU more? It already hits 60c under max load. I might consider a 6 core CPU, but that might be ridiculous and not yield any significant benefit. I heard BF3 fully uses 8 threads so I duno, maybe it would help.

You did it right: Check if your fps are tied to CPU frequency, there is no other way. Btw CPU usage as shown by windows explorer is not entirely reliable. It may show each single thread well below 100% and fps could still scale with CPU speed -> inter-thread dependencies.

Especially on large maps and highly populated servers the CPU will most likely be the issue.

Your CPU is not the issue though. BF3 is one of the worst, if not *THE* worst game you'll ever play when it comes to scaling in cross fire or SLi, it just plain sucks. Perhaps CPU physics play a part but you see the same issue on low end SLI/CrossFire setups as well so it's definitely not entirely your CPU. It's the game.

Proof? Sounds very unlikely. If during CPU physics intensive scenes, GPU load and fps go down, that IS a CPU bottleneck, albeit a temporary one until the player moves away from the action and the CPU gets some breathing room.

 

[soccerballtux]

I have shown you this before. BF3 MP in large open maps 64 player servers is highly CPU bound. It can use 6-8 threads easily.

wow that's incredibly impressive. Who knew Bulldozer could hold its own in well threaded games? It's not even overclocked. Why haven't we seen this before anywhere???

 

[soccerballtux]

Correct. FPS doesn't change at that intense location between 1080p totally maxed with 4x MSAA and 720p no AA.

It seems my FPS will never EVER drop far below 60, occasion drops into 50's. No reason to complain, but i'm not done. Below are posted benchmarks from Oman to see SLI benefit with real numbers and to see if minimums are any better with SLI despite the CPU bottleneck. Something special about these runs; they are worst case with chopper blasting rockets right down on me and screen filled with smoke and total chaos with 60-64 players at construction site area. I had the luck of the chopper action being very similar for both runs. This is about the most CPU intense actual gameplay scenario I could find apart from standing in one spot on the hotel or TV station. This reflects CPU limited situations of actual gameplay.

Benches @ 4.6ghz HT on (will do HT off later)

SINGLE CARD
MIN: 38
MAX: 128
AVERAGE: 80

SLI
MIN: 51
MAX: 178
AVERAGE: 103

This is roughly a 25% improvement in a very cpu limited situation. I'll take it. The difference between 38 and 51 is enough to make gameplay much smoother thanks to the additional GPU headroom, but with a faster CPU it could be much better of course.

100% certain HT off is going to hurt you.
Can you hit 4.7-4.8ghz? That would probably be enough to keep you at 60fps.

 

[Prey2big]

wow that's incredibly impressive. Who knew Bulldozer could hold its own in well threaded games? It's not even overclocked. Why haven't we seen this before anywhere???

It's not a reliable benchmark, it's a very rough estimate. No single fraps run is like the other, at all.

 

[moonbogg]

It's not a reliable benchmark, it's a very rough estimate. No single fraps run is like the other, at all.

That doesn't mean there isn't value in doing it. It plainly shows people what to expect during online gameplay. People understand the variable nature of online play and can still use the benchmarks as valuable info for deciding upgrades etc. OR, you can just buy the best stuff and be done with it, lol.

 

[soccerballtux]

It's not a reliable benchmark, it's a very rough estimate. No single fraps run is like the other, at all.

yea but you can't wave your hands and say that Bulldozer's 2x performance of 2500k means nothing.
So then we just repeat it 4x and average the results and realize there's something to it.

 

[toyota]

wow that's incredibly impressive. Who knew Bulldozer could hold its own in well threaded games? It's not even overclocked. Why haven't we seen this before anywhere???

those numbers seem very odd. why is just a 18% oc on the 2500k giving it about a 40-50% average increase in framerate with spikes even going over 100% increase?

 

[Puppies04]

those numbers seem very odd. why is just a 18% oc on the 2500k giving it about a 40-50% average increase in framerate with spikes even going over 100% increase?

I asked about this data in another thread and AtenRa did try to replicate the same scenario on each run but there are just too many variables to draw any conclusions as far as I am concerned. I see no explanation why an I7 920 would be running at double the FPS (at times) of a 2500K with them both clocked at 4ghz.

I am also sceptical about an 8150 showing an almost constant 20FPS lead over a stock 2500K when just about every gaming bench test anand does shows them either level or the 2500K beating the 8150 by a comfortable margin. I know BF3 can push CPUs hard and the extra "cores" on an 8150 might give it an advantage in certain situations but that graph purports to show the FX chip having up to a 100% lead over the I5 which doesn't bear any resemblance to any other data I have seen even on the most "red" of sites.

It is a nice experiment but at the end of the day all it shows is 4 different runs on the same map in BF3.

 

[BD231]

Proof? Sounds very unlikely. If during CPU physics intensive scenes, GPU load and fps go down, that IS a CPU bottleneck, albeit a temporary one until the player moves away from the action and the CPU gets some breathing room.

Its a software issue just like SC2. For some reason just because it's BF3 the same rules don't apply.

 

[soccerballtux]

I asked about this data in another thread and AtenRa did try to replicate the same scenario on each run but there are just too many variables to draw any conclusions as far as I am concerned. I see no explanation why an I7 920 would be running at double the FPS (at times) of a 2500K with them both clocked at 4ghz.

I am also sceptical about an 8150 showing an almost constant 20FPS lead over a stock 2500K when just about every gaming bench test anand does shows them either level or the 2500K beating the 8150 by a comfortable margin. I know BF3 can push CPUs hard and the extra "cores" on an 8150 might give it an advantage in certain situations but that graph purports to show the FX chip having up to a 100% lead over the I5 which doesn't bear any resemblance to any other data I have seen even on the most "red" of sites.

It is a nice experiment but at the end of the day all it shows is 4 different runs on the same map in BF3.

anand never did 64 players multi though.

I wish someone would just log these numbers for the entire match on their system for 20 matches, average them, and we could put this to rest...

 

[Tempered81]

BF3 Multiplayer benefits from higher minimum framerates with HT, 6-cores, high speed and more cache. A 5ghz 3960X might have a minimum of 80-90, while a 4.5ghz 2500K would produce 55 in the exact same scene.

When the CPU becomes the bottleneck the GPU usage drops. Stare at a flat polygon or mesh in the corner of the map and watch the GPUs peg at 99% 200 fps, then turn around and view the entire map & entities, and witness GPU usage drop to 50/60% and framerate drop to 55 FPS because the CPU is bottlenecked.

BF3 Multiplayer is one of the only instances where a 3960X has a huge advantage in minimum framerates. BF3 is the newest most taxing & stressful multiplayer game for PC hardware. I"ll add the benchmarks here when I find them.

EDIT: Found it courtesy of Grooveriding here at AT

Gents, if you've played BF3 multiplayer, you should know how this works. It's heavily multi-threaded and it's heavily dependent on CPU.

 

[2is]

I don't doubt the extra threads help but I'm curious where you're getting these minimum frame figures from?

 

[Tempered81]

Yeah I edited it, see the 6 core barely breaking below 80 FPS one time, while the 4ghz quad is regularly going into the 50's. That's just counting the CPU difference. Granted it is an old nehalem quad. I'd wager a 2500k versus 3960X would look similar.


Now if you were to bench a single player map, the framerates would be nearly identical.

 

[Don Karnage]

BF3 Multiplayer benefits from higher minimum framerates with HT, 6-cores, high speed and more cache. A 5ghz 3960X might have a minimum of 80-90, while a 4.5ghz 2500K would produce 55 in the exact same scene.

When the CPU becomes the bottleneck the GPU usage drops. Stare at a flat polygon or mesh in the corner of the map and watch the GPUs peg at 99% 200 fps, then turn around and view the entire map & entities, and witness GPU usage drop to 50/60% and framerate drop to 55 FPS because the CPU is bottlenecked.

BF3 Multiplayer is one of the only instances where a 3960X has a huge advantage in minimum framerates. BF3 is the newest most taxing & stressful multiplayer game for PC hardware. I"ll add the benchmarks here when I find them.

EDIT: Found it courtesy of Grooveriding here at AT

Nehalem vs Sandy Bridge Architecture's. Need a 2600K vs a 3930K to compare accurately. You also need to clock them the same.

 

[2is]

Yeah hardly apples to apples comparison.

 

[moonbogg]

I'd love for someone to jump in with a 6 core sandy bridge to help us figure this out once and for all. I am also skeptical about comparing a lower clocked nehalem to a higher clocked sandy. I bet the difference would be pretty big between a 2600k and a 920 when measuring like that.

 

[skipsneeky2]

Like the Op,my dual gtx670 are even held back big time by my 3770k at 1080p.

All same settings as Op but motion blur disabled and post aa on low and i went from 65 fps minimum on 64 player Caspian Border to 71 fps minimum when i overclocked the cpu from 4.2ghz to 4.5ghz.

Enabling post aa to high yields no performance lost nor do i gain any performance disabling it or putting it on low.

I find in my testing that getting into a helicopter and when it rises and picks up elevation,this is one of the few times i find my minimum frame rate as the whole map just pans into view and the fps tank for a few seconds.

 

[Puppies04]

I find in my testing that getting into a helicopter and when it rises and picks up elevation,this is one of the few times i find my minimum frame rate as the whole map just pans into view and the fps tank for a few seconds.

Could this not be also attributed to your GPUs having to repopulate their memory buffers?

 

[moonbogg]

Like the Op,my dual gtx670 are even held back big time by my 3770k at 1080p.

All same settings as Op but motion blur disabled and post aa on low and i went from 65 fps minimum on 64 player Caspian Border to 71 fps minimum when i overclocked the cpu from 4.2ghz to 4.5ghz.

Enabling post aa to high yields no performance lost nor do i gain any performance disabling it or putting it on low.

I find in my testing that getting into a helicopter and when it rises and picks up elevation,this is one of the few times i find my minimum frame rate as the whole map just pans into view and the fps tank for a few seconds.

Could you test your minimum FPS on the sharki TV station looking over the map? I get like 55 up there regardless of lowering res. I'm curious as to how Ivy handles it.

 

[computerparts1]

Looks like you would benfit from upgrading to socket 2011.

 

[skipsneeky2]

Could this not be also attributed to your GPUs having to repopulate their memory buffers?

Could be but it hasn't dipped under 71fps since i clocked my cpu to 4.5 ghz so i doubt its a memory buffer thing,but i could be wrong.

Op my previous post was to suggest upping the clocks on the 2600k as it has made a difference for me and i doubt ivy would make hardly any difference to even consider purchasing.

Ivy is to hot for my liking,so 4.5ghz is as comfortable as i would like to push it on air,a 2600k should yield better overclocking headroom but i certainly hope a patch or driver can help with improving sli.