BF3 CPU bottleneck? Really?

[skipsneeky2]

Wow moonbogg,i hope your going to return your 3930k as that just sucks but honestly i would love a 3930k set up,idk if its drivers but when my cpu hits 40% or higher usage,my game feels a little stuttery and my fps would drop,figured its a cpu memory bandwidth problem.

You experiencing anything sorta like that Moonbogg or did you and not anymore?

 

[moonbogg]

Wow moonbogg,i hope your going to return your 3930k as that just sucks but honestly i would love a 3930k set up,idk if its drivers but when my cpu hits 40% or higher usage,my game feels a little stuttery and my fps would drop,figured its a cpu memory bandwidth problem.

You experiencing anything sorta like that Moonbogg or did you and not anymore?

Yeah, it actually does suck that BF3 didn't respond very much to the extra cores. The X79 is nice to have for other reasons though (need to convince myself of that since i'm too lazy to uninstall everything).
On a more serious note though, its nice to know I have a platform that can handle the next Kepler cards without bottlenecking them and 3930K is just beast and fun to play with. So, I could return it, but by the time I sell my 2600K rig (already got bids on ebay), this new stuff won't end up being that expensive really. If nothing else, it was fun to do the experiment (which im not done with)

 

[MoDeMK]

interesting thread ...

i have some experiences on that too. I was able to test a 2600k@4,4 and a 3770k@4,7 as well as DDR with 1600 and 2133Mhz.

The base are two GTX680 SLI.

Concerning min.FPS on 64 b2k Conquest (gulf of oman) i can say that the upgrade to ivy helped. I was able to achieve min fps of ~85 fps. the 2600k was not able to do that.

also i experienced that the faster memory helped too. Not as much as the ivy CPU but you were able to get the min fps around +5

still i think it is crazy that even a 3770k with those clock speeds limits in multiplayer matches.

 

[skipsneeky2]

Might take Haswell to clear some bottlenecking up but heck since i'm crazy enough to still run a 60hz monitor,the minimums we all pull are perfectly playable but then i could understand the need for a higher minimum if someone was considering 3d.

A 3770k at 4.7ghz is pretty extreme MoDeMK,i had mine at 4.5 and the b4stard blue screened in a middle of a BF3 round which i was dominating in yesterday,its back down 4.2 but at 1.34v at 4.5 load,it was a hot stubborn b*tch for sure.

 

[cytg111]

Results:
min:59
max:166
average:96

EDIT: Just OCed to 4.6 and did a similar run with these results
min:66
max:179
average:103

- So you overclocked ~7% and average increased ~7%.. I'd say that is pretty good proof that you are cpu constrained. In fact it is right there isnt it? staring you right in the friggin face like the eye of sauron.

edit : read the rest of the thread, guess it is worth noting that you are err.. still are cpu constrained, just not cpu's constrained. must have moar clocks.

edit2 : I play bf3 64player maps in the rig below just fine .. but if I have a browser open, I have flash running (ads and all, even with noscript), and flash, for some reason will completely destroy the experience (I use firefox, and will just kill the plugin container(s) before starting).

 

[moonbogg]

Yeah I have to accept that no modern tech will give me what I want. BF3 has always been perfectly playable and runs like glass. I just don't like going from 100 to 55fps. I'd love mins in the 80s but not even an ivy @ 5.0 will give me the 65% increase I would need lol. So screw it I guess. Expensive lesson learned.
In case anyone else was considering a 3930k to help in BF3 this will show them that despite info going around that bf3 uses more cores to good effect in multiplayer, that info is demonstrably false and will lose you money.

 

[2is]

You can turn adaptave vsync on and it will stabalize your frame rate at 60

 

[MoDeMK]

Might take Haswell to clear some bottlenecking up but heck since i'm crazy enough to still run a 60hz monitor,the minimums we all pull are perfectly playable but then i could understand the need for a higher minimum if someone was considering 3d.

A 3770k at 4.7ghz is pretty extreme MoDeMK,i had mine at 4.5 and the b4stard blue screened in a middle of a BF3 round which i was dominating in yesterday,its back down 4.2 but at 1.34v at 4.5 load,it was a hot stubborn b*tch for sure.

yes it is pretty extreme and requires decent cooling running at 1,36v
but it is manageable and seems stable (3-4 days testing).

 

[Rvenger]

Yeah I have to accept that no modern tech will give me what I want. BF3 has always been perfectly playable and runs like glass. I just don't like going from 100 to 55fps. I'd love mins in the 80s but not even an ivy @ 5.0 will give me the 65% increase I would need lol. So screw it I guess. Expensive lesson learned.
In case anyone else was considering a 3930k to help in BF3 this will show them that despite info going around that bf3 uses more cores to good effect in multiplayer, that info is demonstrably false and will lose you money.

It could just be BF3 itself. I don't see any other game behaving in this manner.

 

[moonbogg]

I have a small update. I played with CPU affinity in task manager, went to Oman with 64 players and hid in a corner. I went from 1 core to 6 with HT off.

1 core = 1fps
2 cores = 20fps
3 cores = 35fps
4 cores = 50fps
5 cores = 53fps
6 cores = 56fps

This is about where the FPS was fluctuating so the numbers are not exact of course, but very close. So, it seems that the advantage diminishes fast beyond 4 cores, but a small advantage is there. I didn't notice it before because it was so small...about 6-7fps at best in CPU limited situations. I will add an update with HT on, but I play with it off since it still causes BF3 to stutter.

 

[Rvenger]

Surprised it even ran on 1 core.

 

[2is]

I have a small update. I played with CPU affinity in task manager, went to Oman with 64 players and hid in a corner. I went from 1 core to 6 with HT off.

1 core = 1fps
2 cores = 20fps
3 cores = 35fps
4 cores = 50fps
5 cores = 53fps
6 cores = 56fps

This is about where the FPS was fluctuating so the numbers are not exact of course, but very close. So, it seems that the advantage diminishes fast beyond 4 cores, but a small advantage is there. I didn't notice it before because it was so small...about 6-7fps at best in CPU limited situations. I will add an update with HT on, but I play with it off since it still causes BF3 to stutter.

When was the last time you played BF3 with HT on? To my understanding this issue was fixed quite some time ago. I know I don't have any issues with HT enabled.

 

[moonbogg]

OK HT results...keep in mind I did my best but MP is pretty hard to get good results...here are results with different thread ammounts.

4 threads = 50fps
6 threads = 57fps
8 threads = 63fps
10 threads = 64fps
12 threads = 64fps

Hmm...certainly not conclusive, but I wanted to share the data anyway. It appears that there MIGHT be small improvements up to 8 threads and start to level off after that. I am still confused because certain parts of the game still get very similar FPS to a 2600K. But then again a 2600K has 8 threads, LOL. Maybe thats why my FPS seems so similar to a 2600K, but perhaps it improved by a few.
And if I watch VERY closely, the HT does sometimes cause small stutters and other times large enough to actually notice. Not a game breaker though really.

 

[Termie]

OK HT results...keep in mind I did my best but MP is pretty hard to get good results...here are results with different thread ammounts.

4 threads = 50fps
6 threads = 57fps
8 threads = 63fps
10 threads = 64fps
12 threads = 64fps

Hmm...certainly not conclusive, but I wanted to share the data anyway. It appears that there MIGHT be small improvements up to 8 threads and start to level off after that. I am still confused because certain parts of the game still get very similar FPS to a 2600K. But then again a 2600K has 8 threads, LOL. Maybe thats why my FPS seems so similar to a 2600K, but perhaps it improved by a few.
And if I watch VERY closely, the HT does sometimes cause small stutters and other times large enough to actually notice. Not a game breaker though really.

Wow, really interesting results. Surprising that you found 6 cores was significantly faster than 4 after not "feeling" that difference.

By the way, in your thread analysis, it would be helpful if you also state how many physical cores. For instance, was "4 threads" actually just two cores with HT on? If so, that would suggest that a dual-core with HT would be roughly equal to a quad-core w/o HT, comparing your previous numbers.

 

[moonbogg]

Wow, really interesting results. Surprising that you found 6 cores was significantly faster than 4 after not "feeling" that difference.

By the way, in your thread analysis, it would be helpful if you also state how many physical cores. For instance, was "4 threads" actually just two cores with HT on? If so, that would suggest that a dual-core with HT would be roughly equal to a quad-core w/o HT, comparing your previous numbers.

Yes, exactly why these results should be taken with a grain of salt. All cores were active. So the tests were done between 6 true cores and a seperate test between 6 cores with HT on. Results might be different if I did a test with 4 cores active vs 6 cores active in bios to really simulate a 4 core vs 6 core CPU in a more true way. I'll get to that soon perhaps. Maybe after my peanut butter and jelly sandwich.

 

[Rvenger]

I knew when HT was off it drags it below 60fps. I noticed that a lot going from a 2500k to a 2700k.

 

[Termie]

I knew when HT was off it drags it below 60fps. I noticed that a lot going from a 2500k to a 2700k.

Would you say then that the 2700k performed better than your current 3570k, despite the relatively similar benchmark performance?

 

[moonbogg]

I'll do some virtual comparisons between 2500K, 2600K and 3930K later tonight by disabing cores and HT in bios. So far though, it seems a 2500K is a little inadequate to provide maximal performance in BF3 multiplayer and this trend will probably continue or worsen for future frostbite 2 games, based on the testing already done.
To my suprise, I haven't found any similar testing done by anyone anywhere. Maybe i'm the only guy on earth who really cares? lawl? Perhaps I have a problem.

 

[Pinky Floyd]

What a great, informative thread.

I'm running an I5 3570k @ 4.5ghz with 2x GTX670 in SLI. My system has 16gb RAM and BF3 is installed to an SSD.

In short, I'm seeing the same type of issues as others.

There are certain parts of maps which just kill the framerate. Changing graphics settings makes no difference here. At these points the GPU usage is low and the CPU usage on each core is sky high. We all know the points. Hotel at Oman overlooking the rest of the map, the top of the TV Station on Sharqi.

Until I found this thread I was a bit confused. I'd read threads where people had said they had ONE gtx 670 and their frame rate (not average!) never dropped below 72fps in 64 player multiplayer maps.

That is obviously just plain wrong, misinformative and can only serve to confuse people who buy these cards and don't see what they expect to see from the little fraps counter in the corner of their screen.

Yes, I see a huge advantage to SLI when playing this game BUT I also see a majorly CPU bound multiplayer game where my 4.5ghz Ivybridge just doesn't feel up to the task, quite frankly.

This is the sort of testing which the tech sites should be doing, but instead, the really useful information is buried in a thread like this (and this thread took me a week of searching to find) and I registered to show my appreciation of the work you guys are doing in this thread.

 

[BD231]

Once we get consoles with better multi-threaded support this kinna stuff won't happen. As long as games continue to run a majority of their tasks on one or two cores you're stuck watching your CPU power go to waste. CPU's are plenty fast enough, it's the game.

 

[Rvenger]

Would you say then that the 2700k performed better than your current 3570k, despite the relatively similar benchmark performance?

It depends actually.. In a multithreaded situation, yes. Singlethreaded? Nope. 🙂

 

[B-Riz]

Would the HT and extra cache of a 2700k or 3820 be an improvement over the 2500k?

I am leaning towards a 2700k for better OC, but the 3820 + board would not cost that much more...

 

[Spjut]

Once we get consoles with better multi-threaded support this kinna stuff won't happen. As long as games continue to run a majority of their tasks on one or two cores you're stuck watching your CPU power go to waste. CPU's are plenty fast enough, it's the game.

There's been some complaints regarding the APIs. It seems like a really big weakness they have currently is the draw calls.
http://www.bit-tech.net/hardware/graphics/2011/03/16/farewell-to-directx/2

I think it's unfair to blame the consoles for multithreaded CPUs being held back. Quite a bunch of the early console games, that later got ported to the PC, did indeed show benefits by having 4 cores instead of 2. And so do many still.

It's still going to take time before we get more than 4 cores as a standard.

 

[moonbogg]

Would the HT and extra cache of a 2700k or 3820 be an improvement over the 2500k?

I am leaning towards a 2700k for better OC, but the 3820 + board would not cost that much more...

I haven't done a true comparison test yet, but from what I have gathered so far, the improvement is minimal, but its still there. In a heavy CPU load scenario, a 2700k or 3820 could keep your min FPS closer to 60 rather than falling to the low 50's or even into the high 40's. My chip falls into the mid 50's at some points. I could easily immagine a 2500K going below 50fps in that case.
Your best bet is to go with a chip with hyperthreading, either one you listed, and clock the piss out of it. Clockspeed is the most important thing. Additional threads is just icing on the cake and helps only a little.

EDIT: You need a lot of GPU headroom for any of this to matter at all, BTW.

 

[B-Riz]

I haven't done a true comparison test yet, but from what I have gathered so far, the improvement is minimal, but its still there. In a heavy CPU load scenario, a 2700k or 3820 could keep your min FPS closer to 60 rather than falling to the low 50's or even into the high 40's. My chip falls into the mid 50's at some points. I could easily immagine a 2500K going below 50fps in that case.
Your best bet is to go with a chip with hyperthreading, either one you listed, and clock the piss out of it. Clockspeed is the most important thing. Additional threads is just icing on the cake and helps only a little.

EDIT: You need a lot of GPU headroom for any of this to matter at all, BTW.

Hmmm, yeah. Being that the 2500k is from early last year, I would think a newer 2700k would clock higher and be more stable, but it is all a crap shoot.

And the monitor is 1680x1050, so high textures with a few bits enabled is prolly not being limited by the cpu at 4.0 Ghz.

I am just feeling really lazy towards trying to get another 800 Mhz out of it that may not be stable.

I suppose 4.4 to 4.7 could happen...