Low CPU utilization & Low FPS

[Razorman 10]

PLEASE study a little more, Arma3 & ALL of it's predecessor's require massive cpu power to run well, this game(simulator) over the years has been the driving factor for my upgrades, the rig in my sig was put together only 3 weeks ago in readyness for this game.

Sad but true for those running mid range cpu's in that you will never see great performance unless running very low res & even then you will be limited/bottlenecked by a slow cpu, study cpu's, it's NOT all about 6 cores & ghz, it's about l2/l3 cache sizes & bandwidth for memory (everyone forgets about memory/bus speeds it seems).

I wish you could all have great performance so i could kill ya asses online ;) But sadly A3 is not the game for you if you are new to the series on a midrange rig/setup, hoping for change but not required here.

Thanks for my 0.02c

save some money & then upgrade/build a rig with the biggest baddest cpu possible for you. (allways Intel ;) )

I stand by my original statement, if you disagree fine, i know wtf i'm talking about, i do not wish to discuss further as knowledge appears to be wasted here, therefore i won't.

Once last comment regarding the post above, A3 uses 2+GB of ram here, everything on very high @ 1920x1200.

[Planetside 1]

ArmA never uses more than 1.5gb for me.

[white 1]

ArmA never uses more than 1.5gb for me.

usually less than 1gb for me, and less than 1gb video aswell.

[antoineflemming 14]

It'll turbo to 3 GHz when only loading 2 cores. should perform quite well in a3, especially for a notebook.

And it does ok most of the time. However, it keeps bottlenecking my GPU, and I notice because now after like 15 minutes of playing my GPU load will drop from 99%.

[simjedi 10]

Using the AmAMark test I found the 3 biggest bottle necks are Particles Quality and both Terrain and Object qualities. Just turning Particles down to Standard gave me a 10fps boost from Very High and a 5fps boost from High on test #3. That's a big difference. I also turned down both Terrain and Object qualities to High and now my GPU is at 99% all the time and my CPU usage went up about 15% on each core.....:)

ETA: My View and Object distance is at 1500 for both set in the ArmA3AlphaProfile.cfg file.

Edited March 28, 2013 by SIMJEDI

[dnk 13]

Well this would be a great example if the majority of the problem was the car. But it's not. The majority of the problem is the engine, hence why we've actually seen some improvement after patches. The local mechanic didn't design the car anyway. The real analogy is buying a supercar that uses a faulty engine made by another manufacturer. While you can't blame everything on the other manufacturer (in this case, BIS), more likely than not, the problem the car is having is due to the faulty engine, and not as much on the car being a supercar.

More like you bought a super-engine (CPU) that needed to use 3 flywheels (8 cores: it's imaginary, k) to get its power to the wheels, then tried putting it in a car/drivetrain that only had 1 flywheel (4 cores). And, yeah, the car also has other issues, like a faulty differential and a clunky gearbox that cause problems for many other people, but then here you are adding to the problem with your exotic engine choice that the car manufacturer never considered something important to design for.

because why is it in ArmA 2 Editor I come from 80+ fps to 20 fps when zooming into a forest on empty chernarus with no AI - no Objects no anything at all? Sth. with that streaming process that utilizes CPU, and must run sequential is maybe the bottleneck.

After testing more with AI, I've found the clearest correlation is between:

GPU memory controller load + FPS

It's nearly 1:1 how these move together, along with the actual GPU utilization, of course.

A LOT more:

These changes happen especially when the AI are fighting actively. As soon as there's a lull in the fighting, the FPS/etc increase (though not to pre-fight levels), and drop when it resumes. But, without AI, when the FPS drops, the GPU stays ~100% while this controller load figure stays with the FPS (though it appears it has a ceiling, and when it hits that it will not increase while the FPS still can).

Here are 2 more charts. The first shows me just standing in a forest, looking around. Here, zooming in causes a large FPS drop, while zooming out does the opposite. It appears that the engine decides which LOD to show based on FOV as well as distance, so this is likely why: zooming out causes LOD shifts to lower detail, while zooming in does the opposite. Moving one's head around while zoomed in seems to really stress the memory use of the GPU by forcing rapid loads of many nearby objects, and also hurts CPU use (perhaps getting hung up from all the texture loading onto the GPU, though VRAM is unaffected (orange line))... The purple line at the top is GPU usage (basically always maxed).

The second is another AI test over Agia Marina. Here, I stay clear of the fighting (and high-res LODs) with the same system settings. I mess with the settings a bit midway. Dropping the resolution has the biggest impact, and lessens the memory controller's load greatly (while having a negative impact on GPU use). For the most part, though, as soon as the AI start fighting, there's a sharp dropoff in controller use and FPS, which bounces back slightly during lulls, but otherwise remains low throughout. When I use the binos here, however, the FPS jumps up at the end (with a huge VRAM usage spike for all the newly loaded mid-res LODs).

Overall, it seems that the CPU usage has both positive and negative correlation with FPS in AI fights. My impression of the data is that running the "combat" AI routines is using up too much of the CPU's resources to properly send the necessary data/instructions to the GPU, however the GPU's memory bandwidth may also be a severely restraining factor when AI is not an issue (my card is old, though, so newer hardware with higher bandwidths and faster GDDR5 memory might not suffer from this specific limitation).

bought SimCity for freaking $80 for his Mac PC. So after downloading the game it appeared to be a .exe file. Apparently the Mac version isn't even made yet!!! You think he should study about how retarded EA is before buying a game?

Anyone that buys something without research is (apologies to your friend) an idiot, basically. Your friend apparently bought a game for a Mac without even seeing if they had a Mac version available. And it's EA's fault! :D You have a very, very warped and lacking idea of "personal buying responsibility" and due diligence. Should I inject another automotive analogy here, too? "caveat emptor", always. And the rest of your post is the typical "lazy/stupid devs" trash that for some reason hasn't gotten you infracted yet.

Has ayone ever seen Arma use more than a 1-1.5 gigs of ram? Not me. Normally it's 750-900 or so.

Usually it's around 2GB system + 800MB VRAM. Tried the startup parameters for -maxmem and -maxvmem (something like that) yet? Edited March 28, 2013 by DNK

[chrisb 196]

Very quick test tonight: Current stable release, came out of the devs version to do this test to compare (results later)..

Standing at the side of a wide furrow, slightly hilly area, 1 unit (me) on the map (details below, bottom), result = 66fps at start, not moving. GPU 99% Msi-afterbuner monitoring..

70 ai placed (not seen) on the other side of the map, result = 61fps at start, not moving. GPU 95%

140 ai placed (not seen) on the other side of the map, result = 54fps at start, not moving. GPU 80%

210 ai placed (not seen) on the other side of the map, result = 41fps at start, not moving. GPU 64%

420 ai placed (not seen) on the other side of the map, result = 23fps at start, not moving. GPU 36%

All ai placed at airbase. I was over the other side of the island.

A road just NW of Kamino Firing Range, I was stood just to the west side of the hook in the road, facing north’ish, not moving..

Will be doing the same test later today with cpu and core usage, probably use 'Playclaw' to monitor cpu/cores..:)

Screenies then too..

[sqb-sma 66]

Very quick test tonight: Current stable release, came out of the devs version to do this test to compare (results later)..

Standing at the side of a wide furrow, slightly hilly area, 1 unit (me) on the map (details below, bottom), result = 66fps at start, not moving. GPU 99% Msi-afterbuner monitoring..

70 ai placed (not seen) on the other side of the map, result = 61fps at start, not moving. GPU 95%

140 ai placed (not seen) on the other side of the map, result = 54fps at start, not moving. GPU 80%

210 ai placed (not seen) on the other side of the map, result = 41fps at start, not moving. GPU 64%

420 ai placed (not seen) on the other side of the map, result = 23fps at start, not moving. GPU 36%

All ai placed at airbase. I was over the other side of the island.

A road just NW of Kamino Firing Range, I was stood just to the west side of the hook in the road, facing north’ish, not moving..

Will be doing the same test later today with cpu and core usage, probably use 'Playclaw' to monitor cpu/cores..:)

Screenies then too..

I wonder what the resulting FPS would be if we valued GPU significantly above AI?

Ninja edit: Don't trust playclaw's GPU overlaw, it often goes to 99%/100% for no real reason, when the card isn't at all stressed.

[dnk 13]

Take away all the AI's weapons... Much smaller impact. I can do 100 AI without guns and get ~25-31FPS (pathing and basic routines). Add the guns in, and it's 10-15. Normal is about 34. The combat routines are definitely tying up the CPU somehow (and it doesn't seem related to whether units are in squads or all individuals). This is also related to utilization numbers, as usual. I'm not sure it's bullets/explosions so much, because even when these aren't occurring (fighting lulls), the difference in FPS is minimal.

How BIS can fix this issue is beyond me. As is why AI threads should hold up the GPU commands when the CPU is only going at 50% (even though it's capable of more, as when the AI first encounter each other there's usually a few-seconds spike up to 100%). Something that doesn't occur at initial "combat activation/engagement" but does subsequently is definitely a culprit, and fixing that could have drastic improvements, if possible. That's where my CPU limitation is coming from primarily.

Also, a note to try disabling GPU optimizations for texture quality (setting nVidia's to "Full Quality" or whatever). That made a significant improvement in my tests. Prerendered frames only changed how smooth the FPS changes were, as one would expect.

[guusert 1]

Anyone that buys something without research is (apologies to your friend) an idiot, basically. Your friend apparently bought a game for a Mac without even seeing if they had a Mac version available. And it's EA's fault! :D You have a very, very warped and lacking idea of "personal buying responsibility" and due diligence. Should I inject another automotive analogy here, too? "caveat emptor", always. And the rest of your post is the typical "lazy/stupid devs" trash that for some reason hasn't gotten you infracted yet.

You are retarded. Just don't reply anymore. You are embarrassing yourself. My friend did of course check if the Mac version was available. He downloaded it on the Mac OS version of Origin so you'd expect it to be a Mac game anyway. It said it was available for Mac OS in the requirements.

[Planetside 1]

Enjoying the CPU utilization, I was killed in PvP because a guy bounced out from a rock within 2 frames and I had no time to stop it, plus I couldn't aim due to the absurd mouse lag.

[chrisb 196]

I wonder what the resulting FPS would be if we valued GPU significantly above AI?

Ninja edit: Don't trust playclaw's GPU overlaw, it often goes to 99%/100% for no real reason, when the card isn't at all stressed.

I use 'Playclaw' for cpu monitoring and msi for gpu, works very well that way.

I would only use 150ai in total per mission, in our group we always cut larger campaigns down into smaller missions, that way your getting great fps and sensible ai, well lets say in A2 they were sensible, still working on the A3 ai..;)

Games running very well, haven't had many problems at all since it came out..

[dnk 13]

My friend did of course check if the Mac version was available. He downloaded it on the Mac OS version of Origin so you'd expect it to be a Mac game anyway. It said it was available for Mac OS in the requirements.

And yet it hadn't been made. Seems like something is amiss in your story...

Anyway, everyone knows that EA is not to be trusted, despite your claims to the opposite (which are odd given how despised EA is by the general gaming community - WHATS YOUR ENDGAME?!!! :p). And this is why I support BIS in these threads, because for all the money and super-utilization EA can put into its games (and I've seen complaints to the contrary in other forums, as with nearly every game ever made, the "under-utilization" issue crops up there too), they treat their customers poorly, care only about profits, and make lowest-common-denominator games while squeezing out every last ounce of quality from every franchise they suck up through their money hole.

When people want to then come in and hold up an EA game as the shining example of what BIS should achieve, it's hard to take them seriously sometimes. Small studios that create niche, unique products and care a lot about their communities (and hire from "within" even) are never going to be able to keep up with EA, Unreal Engine, Crytek, and the rest in engine performance, no, but you're getting something special with their products that you can't get from another mindless twitch MP game. Also, I'm still not seeing where BF3 does half the things A3 does in terms of map sizes, draw distances, and AI. /rant

[Sevenz 10]

Yeah, games are generally less optimized for AMD CPU's, but most games don't even use the CPU that much. And if they do, they do it well.

I know few games that are heavy on cpu but i don't know much game who "fully" use a quadcore or more except bf3 and crysis3.

Aion,guild wars 2,wow and assassin creed 3 all of them are big titles who are cpu bound and yet they have issues with cpu/performance.

I wanted too clarify things because you seem to think that making a multithreaded game "fully" using a quadcore or more is easy while in fact there are no games out there doing it and the closest i know are bf3 and crysis 3 so that make two games.

Now you can prove me wrong and give me names of all the other games you're talking about doing it.

[white 1]

I know few games that are heavy on cpu but i don't know much game who "fully" use a quadcore or more except bf3 and crysis3.

Aion,guild wars 2,wow and assassin creed 3 all of them are big titles who are cpu bound and yet they have issues with cpu/performance.

I wanted too clarify things because you seem to think that making a multithreaded game "fully" using a quadcore or more is easy while in fact there are no games out there doing it and the closest i know are bf3 and crysis 3 so that make two games.

Now you can prove me wrong and give me names of all the other games you're talking about doing it.

the ones that become barely or unplayable due to fps and arent multithdreaded, specially nowdays, are the exception.

i have no problem with a game not being multithreaded, as long it performs well with mid-high not-overclocked rigs. but when it doesnt, ppl keep saying that its because its a mil sim and its cpu intensive and to me thats a lie, it just uses the equivalent of 2 cores while the first core bottlenecks everything because of 1 big thread, yeah its intensive, in 1 core. that shows bad design choices and poor execution.

and i agree, that happens, but those are the exception.

http://gamegpu.ru/images/stories/Test_GPU/Action/ARMA%20III%20Alpha/test/arma%203%20proz%20amd.jpg (120 kB)

im not a long time programmer to know about code/game design, but i know a few and asked them, and still even when they dont know each other, their answers were the same about what the game does and how it performs: "bad design" & "bad coding", "dont expect much". sure it could be fixed with enough effort, but since only vague "optimizations" are announced im not sure they will. and here comes bf4 a decade ahead graphicswise and will probably run with at least twice the minimum fps that arma 2 has.

and im done about this issue, ive posted enough about it. if the game launches broken like this ill fight for my money back. and in the rare chance they do fix it, ill buy at least 5 more copies for my friends. (amongst the ones i told not to buy the game as it is now, and they didnt)

Edited March 28, 2013 by white

[dnk 13]

The game is multithreaded. Quit spreading lies because you're angry over your exotic hardware not functioning properly.

[simjedi 10]

The game is multithreaded.

Sure is.

I done some A.I. test with 200 of them at the Airfield for 10 minutes. I set all graphical settings to off or their lowest allowed to get visual processing to the minimum. I went up north to Grid 039070 where it is pretty barren to remove any kind of geometry like trees and buildings also and sat on the ground with nothing but the sky in my FOV.

The first picture has all my cores running at about 50%-70% each.

I repeated this on a second test, but this time I put a frame rate limit of 29fps to limit graphics processing even more. This time though Core #0 shot up to 90% with the other three at about 40%-50% each.

The fps for both was about the same with the limited one being much smoother. I didn't try any higher of a limiter to see if I could have gotten even more fps out of it, but if someone wants to try it go ahead. Maybe someone who gets crappy fps can use a limiter to bring it up and find their balance since the CPU has to hand off less to the GPU.

Edited March 30, 2013 by SIMJEDI

[asd123 1]

Try 1500 viewdistance, 1000 object distance, pip (picture in picture) off.

Your gpu shouldn't be an issue unless you run with aa on with a lot of trees around.

Its about 30-50 fps this way, but it shouldn't it still run better on about 2200, 1800 ?

[Leon86 13]

what mission?

[tet5uo 4]

Sure is.

I done some A.I. test with 100 of them at the Airfield for 10 minutes. I set all graphical settings to off or their lowest allowed to get visual processing to the minimum. I went up north to Grid 039070 where it is pretty barren to remove any kind of geometry like trees and buildings also and sat on the ground with nothing but the sky in my FOV.

The first picture has all my cores running at about 50%-70% each.

http://i.imgur.com/1KbgmCi.png

I repeated this on a second test, but this time I put a frame rate limit of 29fps to limit graphics processing even more. This time though Core #0 shot up to 90% with the other three at about 40%-50% each.

http://i.imgur.com/4jMyIMM.png

The fps for both was about the same with the limited one being much smoother. I didn't try any higher of a limiter to see if I could have gotten even more fps out of it, but if someone wants to try it go ahead. Maybe someone who gets crappy fps can use a limiter to bring it up and find their balance since the CPU has to hand off less to the GPU.

So all we have to do to get somewhat good CPU usage is add a low FPS limit, turn all settings to lowest and stare at the sky making sure we're not near anything with too many polys?

Sounds legit. :D

I will test out this scenario later too.

[gammadust 12]

But do test it... it is academic and unfit for regular arma playing, but it might indeed prove your system does benefit from a fps limiter...

[daze23 1]

so... no word from Dwarden on the analysis?

[Leon86 13]

what analysis?

[daze23 1]

this analysis:

let us first analyze and fix it :) then we will see/talk more ;)

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

[simjedi 10]

I will test out this scenario later too.

Did some more test and it looks like my CPU with running A.I. is best balanced at 35fps. Core #0 is up and even though there is a little bit more usage out of 1-3 at 40fps they stall out less. So until BiS can work on some optimizations I will be using a 35fps limiter for missions with a lot of A.I....:eek:

I mostly play Blitz though so I get a solid 50fps-60fps wherever I go.....:)

Edited March 30, 2013 by SIMJEDI