GPU usage = 0% in A2 Benchmark 02

[Qxs 10]

EDIT: Just another , hey look CPU is bottlenecking and GPU has nothing to do - thread /end-edit

I downloaded Combined Operations from Sprocket.

Installed A2, then OA into the same directory.

It autoupdated to 1.07.

If I run the Arma 2 link, then set my video settings up, and enter the "Benchmark 02" mission, there will be a spike in GPU usage for the first few seconds, then GPU usage drops to 0% (flatline in AMD OverDrive & 0% ATI Overdrive). Fraps measures fps to be about 10 at this point in time.

It will remain at 0% until you get about 45 seconds in, and then it will jump up to about 30%, around the same time you see an increase in fps from ~10 to about 15-20.

To check GPU usage I alt-tab to the desktop (Windows 7 64bit). Where I have left the AMD and ATI utils running along with the Task Manager.

I can reproduce this with little variation in results.

If needed I could probably get the dxdiag info... basically Video settings make no idfference, any settings, same problem.

Video Settings: View Distance ~5000

Low or Very High

Low, High, or Very High

Disabled or Normal

Disabled or Normal

Very Low or High

ANY

Disabled or Normal (High and Very High will bump up GPU usage, I assume this is because it is being moved from the CPU to GPU as previously posted on these forums)

Disabled or Very High

Normal

5:4

-twice I had a strange bluriness for a split second, this effect looks a lot like when you hit ESC and the menu comes up... which I find interesting due to the other thread I read about the bringing the menu up was causing a boost in fps.... but it could have been some strange effect of a missile being fired into my viewpoint or something. -this happens when the helicopter crashes into the 2 large hangers.

-no effect: cpuCount=4 -exThreads=7

-no effect: run as administrator

-no effect: exiting fraps, task manager, AMD Overdrive, ATI Overdrive

-no effect: exit comepletely, and load A2 again.

-no effect: underclocking the GPU

So basically, the video card is only being used for Shadows and possibly Post Processing ( I forget now too many test runs).

Not tested: tinkering with V-Sync, hdr setting, benchmark 01 & 08 under same settings.

EDIT: I did some tinkering with the view distance, and this seems to be a critical setting... GPU usage will actually go UP as you DECREASE view distance, and vice versa... I assume this is because the CPU workload is decreasing and it is therefore able to send more stuff to the GPU (that it was previously just dumping into the trash)... otherwise, I wouldn't expect GPU usage to increase with a lower view distance...

-I also tried the benchmark 02 mission and got some similar strangeness with the view distance, the original problem seems to persist with this mission but it is not as apparent...

-CPU usage remains fairly stable, 50-80% usage, very spikey though...

-RAM usage could be linked somehow, seems to reach a RAM plateau at 1.48GB about the same time the GPU bumps up to ~30%, my gut tells me this is a coincidence.

/end-edit

Edited August 7, 2010 by Qxs
few more details on the end..

[Qxs 10]

I installed ATI Tray Tools, and setup a profile for OA to force V-Sync off. Doesn't seem to change anything (tried on/off).

Did some testing in OA, the 7th Cav Domi West server.

My fps are always ~25... I get about 35 fps with view distance 500, 30 at 1000,

or I can get 25 fps with view distance 5000.

I can 3D resolution to 200%, and get about 15 fps, or I can set it to 50% and get ~25 fps.

Changing the rest of the video settings does nothing but modify the GPU usage a little bit....

My next step is too look at previous driver revisions, but I dunno if I have the burning desire to try to fix something that probably can't be fixed by me.

-no effect: hdrprecision, framesahead...

Also, -cpuCount 2 -exThreads=1 is ignored, and it runs on 4 cores...

Edited August 7, 2010 by Qxs

[Vixente 10]

Why would you underclock your CPU ?

[seelix 10]

downclocking the GPU can prevent CTDs for some users.

(not working on me)

Edited August 8, 2010 by seelix

[Vixente 10]

Ok, didn't know that one. I have similar hardware like the OP. Phenom II 955BE with Toxic HD5850. Both things overclocked, and normally i don't have issues with this game, no performance issues i mean. CPU and GPU use seem normal to me, as i've checked OA uses more GPU than Arma 2 did. I've got W7 64 bits and Catalyst 10.7. I use always the latest functional OA beta.

I did have some strange freezes (only the game) a while back, but i'm not playing so much this days to confirm that they're still happening, but in other areas my computer is 100% stable, this game and all others run smoothly. I never had problems with performance with OA. I have the CPU overclocked @ 3.6 Ghz with default vcore, i guess yours should be capable of similar OC, or maybe 3,7 or 3,8... To assure stability run LinX for 1 hour and FurMark for your GPU. If they pass the test, then your computer is rock-solid.

This are mi settings:

Edited August 8, 2010 by Vixente

[Qxs 10]

Vix, what do you get in the 3 benchmark missions, running them from OA.

Same settings as that screenshot would be fine...

[Dead3yez 0]

Same CPU. (965) Worse intels are getting better performance it seems sometimes.

http://forums.bistudio.com/showthread.php?t=91860

[Vixente 10]

First, i normally don't benchmark, only when i get a new piece of relevant hardware. I don't know if my results are adequate or not for my hardware. You tell me.

- All in-game settings like my previous screenshot. (I don't change them, ever.)

- Sapphire Toxic 5850 overclocked @ 860/1300

- Arma2 folder in a different drive than the pagefile.sys

- 6 GB RAM (meaningless... as the game barley uses 2gb) DDR2 1066 Mhz.

- OA beta patch 72418

Chernarus Benchmark 1: 38 FPS

CPU & GPU usage:

Comment: This first bench seems equilibrated.

Chernarus Benchmark 2: 15 FPS

CPU & GPU usage:

Comment: Very heavy benchmark, looking at the graphs i assume it is way heavy on the CPU. GPU barely goes up. I think i get here the same as you. Probably this bench is screwed up somehow. No idea.

Takistan Benchmark: 44 FPS

CPU & GPU usage:

Comment: This is what a normal PC Game should be, very heavy on the GPU, and "normal" use on the CPU.

I do know that most software don't tend to force the CPU at maximum. I am no hardware geek, but it seems that the CPU does only go near 90% or 100% when using heavy calculating programs, like LinX or other specific CPU benchs. I see every time this kind of "balancing" with the 4 cores in a lot more games appart from Arma.

I really don't have a clue if my values are good, bad or average. Don't really care, because my final experience with the game is good. No laggy graphics, all very smooth, very playable. Like all other games, all this tend to go down when more units and AI are near you. But this is normal, right? :rolleyes:

Sorry for my poor english btw.

EDIT: OK, i've read all your post Dead3yez, and now i can understand the problem with Chernarus benchmark 2. It is sad that no one from BIS replied to your post months ago :(

Edited August 9, 2010 by Vixente

[Qxs 10]

EDIT: After reviewing the AI problems/Deadeye posts/threads, it would seem like this is the issue with this benchmark

Chernarus Benchmark 2: 15 FPS

CPU & GPU usage:

Comment: Very heavy benchmark, looking at the graphs i assume it is way heavy on the CPU. GPU barely goes up. I think i get here the same as you. Probably this bench is screwed up somehow. No idea.

EDIT: OK, i've read all your post Dead3yez, and now i can understand the problem with Chernarus benchmark 2. It is sad that no one from BIS replied to your post months ago :(

Unless this mission is specifically engineered to perform differently than a typical mission (false representation, which seems unlikely), than any problem with this mission is likely going to be a problem for various other missions, be it multiplayer or whatever...

My point of view is that the problem is right here in the benchmark mission, you don't need to look anywhere else. I will proceed with testing various catalyst versions as soon as I get some time to do so.

Updated/New Data:

Changes:

SuperPreFetch=1, boot files only, but I see it is still caching armaoa.exe, so it's probably not actually changed...

Added framesahead and hdr to the profile and marked it read-only.

Setup:

-Task Manager open and watching CPU usage

-ATI Tray Tools, GPU % running

-Fraps running (may not have been running for all of them)

-Alt tab after mission ends to export the graph

-AMD OverDrive is run to adjust CPU Multiplier, applied, and then exited each time.

-Started at 3.2ghz, then 2.6, then 3.4, and finally 3.4 again while bumping shadows up to Very High (to see the effect of switching it from CPU to GPU usage.

-Video Settings located within

-My Game Profile, if needed, located within -broken file atm

OA Benchmark 02 (Ati Tools GPU Graph - Exported):

my Tripod website with jpeg links

3.2ghz = 16, 15, 14 fps *I ran it three times in a row because there was massive HD activity the first run, and considerably less the second time*

2.6ghz = 12 fps

3.4ghz = 15 fps

3.4ghz, Shadows VH = 15 fps

Conclusion:

The shit is messed up.

You can see some dependence on CPU clock speed, but honestly, 2 or 3 fps difference between 2.6Ghz and 3.4Ghz, on 4 cores no less, sounds ridiculous to me....

@3.2Ghz the GPU is pegged for about 25 seconds... what on earth could possibly be that much different that it needs 3x the usage it had before, and the same time frame....

I will look at the deadeye Dead3yez stuff after work.. .

EDIT: After reviewing the AI problems/Dead3yez posts/threads, it would seem like this is the issue with this benchmark. I will move on to finding out what bushes kill my fps with rifle scopes instead... :/

Edited August 10, 2010 by Qxs

[Vixente 10]

I think you can "play" any time you want, you are not going to improve the result with that benchmark. If you have read Dead3yez, there is a very interesting experiment, and the conclusions is that there is something wrong with how the game processes AI, and how the game manages its threads with the CPU cores too. That Benchmark has a lot of IA, you can see tanks, infantry, airplanes, i don't know how many there are, but certainly there are not a few. But looking at the GPU graph it's clear that the GPU has no time to work its shit. And seeing your tests (and others), overclocking or other tweaks have nothing to improve because of "that issue" with the AI/CPU thing.

Take a look at Dead3yez post if you haven't yet. It's worth the reading.

[zaira 10]

EDIT: After reviewing the AI problems/Deadeye posts/threads, it would seem like this is the issue with this benchmark

Unless this mission is specifically engineered to perform differently than a typical mission (false representation, which seems unlikely), than any problem with this mission is likely going to be a problem for various other missions, be it multiplayer or whatever...

My point of view is that the problem is right here in the benchmark mission, you don't need to look anywhere else. I will proceed with testing various catalyst versions as soon as I get some time to do so.

Updated/New Data:

Changes:

SuperPreFetch=1, boot files only, but I see it is still caching armaoa.exe, so it's probably not actually changed...

Added framesahead and hdr to the profile and marked it read-only.

Setup:

-Task Manager open and watching CPU usage

-ATI Tray Tools, GPU % running

-Fraps running (may not have been running for all of them)

-Alt tab after mission ends to export the graph

-AMD OverDrive is run to adjust CPU Multiplier, applied, and then exited each time.

-Started at 3.2ghz, then 2.6, then 3.4, and finally 3.4 again while bumping shadows up to Very High (to see the effect of switching it from CPU to GPU usage.

-Video Settings located within

-My Game Profile, if needed, located within -broken file atm

OA Benchmark 02 (Ati Tools GPU Graph - Exported):

my Tripod website with jpeg links

3.2ghz = 16, 15, 14 fps *I ran it three times in a row because there was massive HD activity the first run, and considerably less the second time*

2.6ghz = 12 fps

3.4ghz = 15 fps

3.4ghz, Shadows VH = 15 fps

Conclusion:

The shit is messed up.

You can see some dependence on CPU clock speed, but honestly, 2 or 3 fps difference between 2.6Ghz and 3.4Ghz, on 4 cores no less, sounds ridiculous to me....

@3.2Ghz the GPU is pegged for about 25 seconds... what on earth could possibly be that much different that it needs 3x the usage it had before, and the same time frame....

I will look at the deadeye Dead3yez stuff after work.. .

EDIT: After reviewing the AI problems/Dead3yez posts/threads, it would seem like this is the issue with this benchmark. I will move on to finding out what bushes kill my fps with rifle scopes instead... :/

...2 or 3 fps on 15 is 20 % difference

benchmark 2 is cpu benchmark (hard on gpu 2).

If u want to properly see that the game engine is cpu bond in that benchmark u obviously test at low resolution and details (like lowest 640, and 3d res 50%).

On my q6600 game scales lineary with cpu clock on that test, implication is cpu bottlenecks.

As i seen for my 1+ years on arma 2 playing, game is on places cpu bottlenecked (like in warfare), gpu bottlenecked (forrests, town), cpu+gpu bottlenecked (manhatan, elektrozazavodsk mission).

If u want to play arma2 like u are probably used to, buy most expensive cpu+gpu combo, and overclock it, even then your fps can drop to sub 30.

But honestly in multi the game is playable with modern cpus and gpus (like i5, i7 and FTX 460+, ati 5850+) at decent visuals.

[Vixente 10]

benchmark 2 is cpu benchmark (hard on gpu 2).

Have you seen our tests and graphs ? GPU does almost nothing in bench 2. And the CPU is not as heavy loaded as you may think.

I bet you that even with i7 980X you will get poor results in bench 2. It's not the hardware, it's the game. Check Dead3ye thread too.

[zaira 10]

Have you seen our tests and graphs ? GPU does almost nothing in bench 2. And the CPU is not as heavy loaded as you may think.

I bet you that even with i7 980X you will get poor results in bench 2. It's not the hardware, it's the game. Check Dead3ye thread too.

Shure i did seen tests, and graphs, and dead3eys tread.

I did my own tests, and cpu is bottleneck in that test.

You can check it for yourself, easly.

Just eliminate possibility that GPU is bottlenck, lower everything in video options, and test it.

And play with clock and cores to see fps change.

Like i get 18fps wih 4x3.2 core 2 quad on 3.2GHz, and i get 20 with same at 3.6GHz 12.5% clock speed up gives me same percentage of fps boost.

What realy is disipointing is that if i run the game with only 2 cores, and force affinity on 2 cores, i only get 15% less, so implication is that game realy isnt optimised for quads on that test.

Arma 2 likes high clock, more than cores, shure there is benefit with 4vs2 cores but not that much that you would call the game 4 core optimised.

I think that AI thread/s kill the performance, as rendering thread needs to wait for AI thread/s to finish. Arma 2 uses 15 threads, (resource manager).

That 0% is rubbish, u cant realy measure gpu, and cpu usage just that simple, as u think, but low usage realy points to my hypotessis.

Well, the sadest part is that i spend more time to analyzing games performance than actualy playing it, and i play it alot and in a active clan:eek:

You are totaly right about 980, not even that cpu cant play game without fps sometimes droping tu sub 30 (and i mean 4+GHz overclock).

Maybe with zambezi and sandy bridge overclock we could play the game without fps drops.

So with any BIS game, u just need to wait 2 years to play it with eye candy ON....sad but true

Edited August 11, 2010 by zaira

[Vixente 10]

Arma 2 likes high clock, more than cores, shure there is benefit with 4vs2 cores but not that much that you would call the game 4 core optimised.

I think that AI thread/s kill the performance, as rendering thread needs to wait for AI thread/s to finish. Arma 2 uses 15 threads, (resource manager).

Exactly, but i don't think we should call it "bottleneck" when the problem is the software optimisation/thread management/core use balancing/whatever.