Low CPU utilization & Low FPS

[Illusion_cz 10]

arma3alpha.cfg , set GPU_MaxFramesAhead=1; and check your fps/smoothness ;)

Well this helped a lot in Arma 2. Game was much smoother and responded better. Not sure If I got more FPS but I realized big change. Lets see what it does in Arma 3.

[daze23 1]

no - i mean in more detail. It was rather a rhetorical question of mine, I wanted to go deeper on that matter with an open mindset.

because as I get it, its just always some kind of approximation to the real thing. But I have really no hardware technical insight.

just an example what I think it works: lets say your CPU has 10 cycles or time frames to do something, and in 9 frames it does nothing, but then there comes this one task and the CPU is at 100%. Now you measure your CPU load and the measurement cannot measure every small frame, but lets say 10 frames. So it will tell you: cpu load is 10% - even if it was 100% at some time and 0% at some other time.

so what basically is needed is some splitting of of tasks into several timing frames? because my cpu has a load of around 30% in ArmA. but when I overclock it, I have double fps and still only 30% load. so when Overclocked those little spikes that need 100% CPU are calculated much faster. But there are still those times when the CPU has idle time. right?

so it is all about is it possible to split the calculations into several timing frames?

And I always thought CPUs have several instructions, like integer operations, floating point, and some other special stuff (descrp. in CPU Manuals). So if we find out what operations / functions are mostly needed in ArmA we could rethink the code thats based for these cals?

the problem is: what if we have 10 tasks and 10 timing frames. but every tasks needs the output of the task before him? so he mustz wait for that prior task, and only then can start its own task.

Normally such complicated tasks and calculations are only for science stuff etc. but maybe herein lies armas problem: the AI?

But then I dont udnerstand why we have even low fps in Multiplayer without AI?

Bascially without looking deep at the code, and having also hardware udnerstatement we cant really tell whats the problem, and if it is even possible to "fix" it right?

to me, "the problem" is my gpu is being held back to 20-30%. whatever my cpu is doing, with however many cores, it is creating a HUGE bottleneck

[jumpinghubert 48]

@Minoza

to disable AA with a GTX670 its a waste of gpu-ressources. Without AA you might have 55fps and 40% gpu-usage in a certain situation. With full AA you might have 55fps and 89% gpu-usage.

the most important thing is to reduce all cpu-intensive graphic settings like object radius, terrain detail and object detail AND second to push all gpu-intensive graphic settings IF the gpu usage is lower than 100%. Thats all for the moment of alpha you can do.

[daze23 1]

@Minoza

to disable AA with a GTX670 its a waste of gpu-ressources. Without AA you might have 55fps and 40% gpu-usage in a certain situation. With full AA you might have 55fps and 89% gpu-usage.

but the fundamental problem there is that your gpu in being bottlenecked. it seems reasonable when you're talking about 55fps, but what you said is also gonna be true at 20fps. basically people are setting their video settings too high, because it doesn't make a difference at this point. the 670 is a good card, but honestly if the game were working properly, I doubt you would want to use 8x AA with it

[white 1]

but the fundamental problem there is that your gpu in being bottlenecked. it seems reasonable when you're talking about 55fps, but what you said is also gonna be true at 20fps. basically people are setting their video settings too high, because it doesn't make a difference at this point. the 670 is a good card, but honestly if the game were working properly, I doubt you would want to use 8x AA with it

you are probably right but i m not sure, i reinstalled bf3 yesterday because of the new end game expansion, im able to sustain 50+fps with 4x msaa and everything on ultra, on fullhd. i have a 660ti oc. 32 player online game. damn thats a beautifull game.

[daze23 1]

you are probably right but i m not sure, i reinstalled bf3 yesterday because of the new end game expansion, im able to sustain 50+fps with 4x msaa and everything on ultra, on fullhd. i have a 660ti oc. 32 player online game. damn thats a beautifull game.

I'm just saying that having 8x AA would pretty much be the last thing I would do. it's very expensive performance wise (in an engine that actually lets your gpu work), and doesn't deliver that much of a visible difference. I'd rather prioritize a higher framerate, and other 'eye candy', before turning up AA to the max

[Dwarden 1124]

What about this : GPU_DetectedFramesAhead=0;

Mine is zero. Should I change it or keep as it is?

ignore detectedFrames, that's written by engine (in OA) but afaik ignored in A3

---------- Post added at 19:27 ---------- Previous post was at 19:24 ----------

- If you have the CPU and the GPU usage both at 100%, ideally, the system is perfectly balanced and the engine is working fine

- If you have the CPU at 30% and the GPU at 100%, then the CPU is bottlenecked by the the GPU

- If you have the GPU at 30% and the CPU at 100%, then the GPU is bottlenecked by the the CPU

- If both are never at 90-100%, then the game engine is broken somewhere which is the case for Arma 3 right now

Plain and simple.

you fail to realize that 100% usage cores don't mean 100% efficiency of the code ... I can just create dozens threads to stall all cores and they will do nothing ... so stop talking in terms of % cpu usage but efficiency of use of MT on multiple native cores/cpus w/e

[WalkerDown 1]

[/color] you fail to realize that 100% usage cores don't mean 100% efficiency of the code ... I can just create dozens threads to stall all cores and they will do nothing ... so stop talking in terms of % cpu usage but efficiency of use of MT on multiple native cores/cpus w/e

Yup, coz BIS is fair, while the other developers are wasting the cpu cycles with fake threads to fake the CPU usage. I'm not sure if it's more invalid his argument (cpu usage = code efficiency) or yours (wasting cpu usage with threads doing nothing). What i know really, is that a game that runs very bad and that barely uses the 30% of the CPU and GPU is surely exposes a very bad coding and optimization, because my car is capable of going much faster than this, but you designed the gas pedal so bad that it cannot be pushed for more than the 30% of its full travelling.

Dwarden, here none is complaining about the current situation (at least none with half brain), because it's alpha, but most of us are complaining about the fact that you (BIS) didn't recognized the problem and didn't make it clear on your SITREP. Until this happens publically we can only believe that you do not consider it the first bug to solve (despite what mantis says) or you have no idea how to fix it. We don't need your personal messages here on the forum, we need the issue to be recognized officially by BIS to make it clear it's being worked on TODAY.

[Dwarden 1124]

we don't recognize the problem ?

the sole fact we are discussing it, means we're aware of 'things that shall be improved' ...

anyway like I said before one of reason the low FPS is linked to some other problems we found in engine ...

after that's solved we can return on the 'efficiency' discussion

+++

btw. just something to think about, are you aware AMD drivers seems to lack any/good enough support for DX11 Multithreading features ?

yes, the technology Civilization 5 developers were so proud about (it forced in 2011 even NVIDIA to add it into drivers)

http://www.rage3d.com/board/showthread.php?t=33976408

now when one realize that thread is being discussed already 2 years, then I wonder why there isn't huge crowd of 'users' demanding it ;)

afaik the MT drivers optimizations exist only for DX9 titles, while for 10/11 they did disabled them or just used minor tweaks

[Insanatrix 0]

ignore detectedFrames, that's written by engine (in OA) but afaik ignored in A3

---------- Post added at 19:27 ---------- Previous post was at 19:24 ----------

you fail to realize that 100% usage cores don't mean 100% efficiency of the code ... I can just create dozens threads to stall all cores and they will do nothing ... so stop talking in terms of % cpu usage but efficiency of use of MT on multiple native cores/cpus w/e

http://en.wikipedia.org/wiki/CPU_time

It's a measurement of work over time, or rather a measurement of indexed performance versus input/output operations. In fact it's more of a measurement of work done per I/O cycle by the CPU. What you are describing is something closer to CPU load in a UNIX environment. Really the same goes for GPU usage as the amount of work done by the CPU and GPU is dependant upon the amount of data fed to it by the I/O sub-systems.

If you're below 99% CPU utilization it means that the CPU is waiting on the I/O sub-system in order to finish it's task, therefor it's completing the task slower than the CPU is capable of doing. All that it means is that the CPU is simply waiting for something to be delivered before it can finish it's task and move on, and this measurement of time happens in nano seconds. It's not a measurement of code efficiency.

Oddly enough this is the point I'm trying to get across in the 64 bit thread, that the I/O system is stalling the CPU threads due to the fact that it is trying to move so much data across such a narrow bandwidth channel ( SATA interface), rather than a proper interface such as the Hypertransport/Memory Bus/System Bus, that it's saturating the channel and causing everything else to be under-utilized while it struggles to keep up.

Edited March 17, 2013 by Insanatrix

[NeuroFunker 11]

@NeuroFunker

Your guide is half good. It points into right direction but so much wrong stuff, let me start:

(I have 670 too, 3770K and 16GB ram)

Tab - BASIC.

Resolution, gamma, brighness, interface size aren't important, lets begin with Visibility:

I can pull 5k overall view distance with fps around 50, which is ok for me and I think most of the people here. I keep object at 1600max, it might sound like big difference from overall rendering distance but in reality it looks great and I find it optimal. I agree on shadows part though.

Overall - i have set to 2,8k (I use 5k)

Object - to 1,8k (I use 1,6)

Shadow - 200m

Next Tab - Rendering:

Rendering Res - 100%

Vsync - enabled (now why would you want to have this enabled with avg. of 30 is beyond me, I always use adaptive vsync, I recommend turning this off for both NV and ATI users unless you get 60+ framerates)

It is not correct that AA and ATOC doesn't affect performance, it kills the performance, just try looking through scope when dense vegetation is in front of you - good luck! Enjoy a slide show.

Antialiasing - 8xAA (wtf?) - I disable AA and use only PPAA

PPAA - FXAA Ultra

AtoC - All trees + grass. - If you want to keep AA (2-4 max advised), disable this unless you enjoy a slide show whenever you aim through scope and dense grass

Postprocess Quality - Normal (I keep this at max setting, it doesn't affect my performance in a way you described)

HDR Quality: Standard

Anisotropic Filtering - Very High (I keep it at Ultra)

PiP (Picture in Picture)- Ultra

Dynamic Lights - Ultra

Next Tab - Quality:

Texture Qualty - Ultra (I keep it at Very high, no visible difference, eats more VRAM for no purpose on Ultra)

Obects Quality - Standard (High)

Terrain Quality - Standard (High)

Clouds Quality - Ultra

Shadows Quality - High

Particles Quality - Very High (I keep it at High, there's a slight difference but much better fps when there's a lot of smoke for example)

So, with my settings, I give you better overall quality and better framerate. Most noticable killers in your settings are in order (higher to lower); Vsync, ATOC(AA), object render distance. You can gain more with pushing Objects and Terrain quality to High and overall view distance to 5km.

Thanks for critics. Like i said, my settings are basics you could tweak on, i can't guarantee it will run same good as by me, but after hours of tweaking, trying to find best balance of quality/performance, behind arma 2 tweaking experience i ended up with this settings.

Like you see, even on similar build it differs. I can't play with both terrain and object details to high, it kills my performance next to view distance, and it doesn't necessary look better if not at all, but what it does, it kills my performance a lot.

Same with post process quality, does it look better on high then normal? Not to much, but it kills my framez as well.

Same thing with 8xAA, compare your fps with off and on, and look at gpu usage. At my rig, it never gets over 50% without AA and it peaks maximum to 90% sometimes, with 8xAA and doesn't kill my performance, but looks way better then without, so why not use it? Slide show? No, i haven't, you might be comparing it to Arma 2/OA, yes there is a slideshow, even without AA when looking at grass/trees through scope, not in arma 3 for me.

Same with Atoc, i doesn't change anything for me, except it looks simply better with it, can you stand that ugly vegetation without atoc? I can't.

Edited March 17, 2013 by NeuroFunker

[CalinTM 1]

So i guess there is still no fps/usage fix ? Any official response ?

[EpidemicVII 1]

Download "Razer Game booster ", it drastically gave a huge change to the FPS, i also used D3OVERDRIVE, I can now kill enemies ...............while loooking at them in the eyes. Turn on sync also, and lower resolution, happy gamings.

[Minoza 11]

Thanks for critics. Like i said, my settings are basics you could tweak on, i can't guarantee it will run same good as by me, but after hours of tweaking, trying to find best balance of quality/performance, behind arma 2 tweaking experience i ended up with this settings.

Like you see, even on similar build it differs. I can't play with both terrain and object details to high, it kills my performance next to view distance, and it doesn't necessary look better if not at all, but what it does, it kills my performance a lot.

Same with post process quality, does it look better on high then normal? Not to much, but it kills my framez as well.

Same thing with 8xAA, compare your fps with off and on, and look at gpu usage. At my rig, it never gets over 50% without AA and it peaks maximum to 90% sometimes, with 8xAA and doesn't kill my performance, but looks way better then without, so why not use it? Slide show? No, i haven't, you might be comparing it to Arma 2/OA, yes there is a slideshow, even without AA when looking at grass/trees through scope, not in arma 3 for me.

Same with Atoc, i doesn't change anything for me, except it looks simply better with it, can you stand that ugly vegetation without atoc? I can't.

Yeah I agree completely with you! Don't get me wrong, I think everyone here appreciates the effort, I just commented from what I've experienced, I might be wrong, or my setup wouldn't work well for someone else with similar rig. Yeah, I don't like how vegetation looks without ATOC but it's a price I have to pay because for me, it brings framerates down from high 50's to high 20's whenever looking through scope and dense grass at the same time. Dunno why, but it is somewhat reassuring to hear it is not happening to you as well, it might be a bug only.

As for AA @JumpingHubert, it's just my preference, I've got enough juice for 2-4 AA, but ATOC is giving me a headache and I don't want good overall AA if grass will look like crap in comparison, I rather have everything covered by PPAA then.

[doveman 7]

Regarding any boost from putting Objects Detail on Standard or High, as with A2 I'm seeing LOD switching on trees (they fill in/thin out as you move your viewpoint) which I find very distracting when trying to keep an eye out for enemy movement. I think putting Objects Detail on V.High might prevent it (with A2 I found this fixed LOD switching on buildings but not on trees) Perhaps there's some other combination of settings that prevents/covers-up this, so perhaps not everyone is seeing this or it just doesn't bother them but I really can't put up with it. Be interested to know if anyone else observes this problem with Objects Detail on lower than V.High.

http://forums.bistudio.com/showthread.php?137726-LOD-switching-issue-since-1-62/page21

[Minoza 11]

Download "Razer Game booster ", it drastically gave a huge change to the FPS, i also used D3OVERDRIVE, I can now kill enemies ...............while loooking at them in the eyes. Turn on sync also, and lower resolution, happy gamings.

This is as misleading as it can get. This software was actually tested and conclusion was no measurable performance gain existed. I'm not saying you didn't get performance gain, but if you did, it means there is something seriously wrong on your side, if this software helps you.

[daze23 1]

This is as misleading as it can get. This software was actually tested and conclusion was no measurable performance gain existed. I'm not saying you didn't get performance gain, but if you did, it means there is something seriously wrong on your side, if this software helps you.

well if there's one game it would help in, it would probably be this one. most games won't benefit much from freeing up some memory and a little cpu usage. but apparently in this game, everything matters:

http://forums.bistudio.com/showthread.php?147724-STEAM-performance-troubles-and-how-mitigate-them

[jumpinghubert 48]

@JumpingHubert, it's just my preference, I've got enough juice for 2-4 AA, but ATOC is giving me a headache and I don't want good overall AA if grass will look like crap in comparison, I rather have everything covered by PPAA then.

hmm...msaa and fxaa have no antialiasing effect for me.

[NeuroFunker 11]

Yeah I agree completely with you! Don't get me wrong, I think everyone here appreciates the effort, I just commented from what I've experienced, I might be wrong, or my setup wouldn't work well for someone else with similar rig. Yeah, I don't like how vegetation looks without ATOC but it's a price I have to pay because for me, it brings framerates down from high 50's to high 20's whenever looking through scope and dense grass at the same time. Dunno why, but it is somewhat reassuring to hear it is not happening to you as well, it might be a bug only.

As for AA @JumpingHubert, it's just my preference, I've got enough juice for 2-4 AA, but ATOC is giving me a headache and I don't want good overall AA if grass will look like crap in comparison, I rather have everything covered by PPAA then.

nah, it's all fine. AS for ATOC killing performance, try turn off post proces off and try turn atoc on, maybe in combo this causing fps drop.

[Minoza 11]

nah, it's all fine. AS for ATOC killing performance, try turn off post proces off and try turn atoc on, maybe in combo this causing fps drop.

Good idea, I'll try that out.

[Dwarden 1124]

I have several AMD users reporting way higher ARMA 3 Alph fps with new AMD 13.3 beta 2 driver

http://support.amd.com/us/kbarticles/Pages/AMDCatalyst13-3WINBetaDriver.aspx

about the efficiency seems like the person failed to understood the difference between bad code which occupy cpu resources and good code which utilizes the occupation properly :)

[tremanarch 6]

http://en.wikipedia.org/wiki/CPU_time

It's a measurement of work over time, or rather a measurement of indexed performance versus input/output operations. In fact it's more of a measurement of work done per I/O cycle by the CPU. What you are describing is something closer to CPU load in a UNIX environment. Really the same goes for GPU usage as the amount of work done by the CPU and GPU is dependant upon the amount of data fed to it by the I/O sub-systems.

If you're below 99% CPU utilization it means that the CPU is waiting on the I/O sub-system in order to finish it's task, therefor it's completing the task slower than the CPU is capable of doing. All that it means is that the CPU is simply waiting for something to be delivered before it can finish it's task and move on, and this measurement of time happens in nano seconds. It's not a measurement of code efficiency.

Oddly enough this is the point I'm trying to get across in the 64 bit thread, that the I/O system is stalling the CPU threads due to the fact that it is trying to move so much data across such a narrow bandwidth channel ( SATA interface), rather than a proper interface such as the Hypertransport/Memory Bus/System Bus, that it's saturating the channel and causing everything else to be under-utilized while it struggles to keep up.

could you elaborate on this one please?

[MaxiM_PL 1]

Unfortunately, even if there is any difference between CAT 31.1 and 13.2b2 in my case it's negligible. ArmA III still uses 40% of my 7870 tops. Quite possibly because the overall usage of my i7-3770k is about 30% when running the game. Hence the conclusion: software bottleneck.

[Dwarden 1124]

Unfortunately, even if there is any difference between CAT 31.1 and 13.2b2 in my case it's negligible. ArmA III still uses 40% of my 7870 tops. Quite possibly because the overall usage of my i7-3770k is about 30% when running the game. Hence the conclusion: software bottleneck.

erm 13.2b2? this is 13.3 beta 2, from this weekend

[stk2008 14]

Dwarden could you share with us this comment please

"anyway like I said before one of reason the low FPS is linked to some other problems we found in engine ...

after that's solved we can return on the 'efficiency' discussion"

What did you find in the engine im just interested please share.

Thanks