ArmA2 Community Issue Tracker

I get this too and so do some friends I play with. The only remedy that works for me is setting model details to low. In Chernogorsk that gives me 30fps, though it drops to about 20 when there's lots of stuff going on. Performance in towns definitely needs to be improved.

This is still the case for 1.02

Issue still as bad with PP disabled?

To attach the Arma2.cfg and arma2profile are a must for performance issues.
Dxdiag as well please.

Problem persists in version 1.02.

I've begun working on a "Chernogorsk benchmark" mission that flies a camera through a pre-defined path through the town and measures FPS (maximum, average, average minimum and absolute minimum). I'll upload it here ASAP. Should be helpful in determining what settings are best to help with performance.

Cheers MDX!

Here's the promised Chernogorsk benchmark mission. It will fly a camera through a set series of positions in Chernogorsk. At the end, you will see an overview of different framerates.

Here's a short explanation of the overview at the end:

- The average framerate is basically your overall average fps during the whole benchmark.
- Average minimum is the average low your framerate dropped to. This will always be lower than the overall average. The lower it is, the more inconsistent your framerate was. If this is very low (10-15fps), it means you probably get a lot of stuttering.
- Highest is, as the name indicates, the highest your framerate ever went during the scene, even if it was just one frame.
- Lowest is the lowest your framerate went. This will probably under 10fps.

Of course for ideal visual quality, all values should be the same. Even if all were on 30fps, it would still mean you got an absolutely rock-solid and stutter-free 30fps, which would be fine, even for gaming. If you got an average of 50fps but an average minumum of 10fps it means your framerate is good overall, but you get a lot of stuttering.

In my first tests I noticed that lowering object detail greatly increases overall performance, but doesn't help much against stuttering. I'll keep at it.

I realized earlier that the benchmark mission I uploaded contains an Intro. You'll need to skip it if you place the mission in your missions folder and run it as a single player scenario. I recommend placing it in the editor instead. I'll upload a new version later.

People test this - share your results!

My HW: Q6600, 4 GB RAM, NV 9800 GT @ 182.50. Intel M SSD.
All setting at normal, shadow and PP off. AF max. No maxmem.

Finally tested this - great test mission!
Please upload the fixed version for more noob friendly use.

It would be nice to show data during the camera scene.

Important to make two runs at least:

Here is my benchmark. I noticed that my HDD is probably the biggest show stopper in this case. Usually I don't get that bad loading lag down to 3 fps (or really rarely), since I'm not "running" as fast the camera goes; maybe the camera speed is slightly too fast?

My specs:
Core i7 920, 3 GB RAM, Nvidia 8800 GTS 512MB @ 186.18
All settings at normal, except shadows: high, graphics mem: high, PP: low

EDIT: kju, whats a NV 9800 PRO? it's gotta be a GT or GTX.

Fixed and added SSD as info.

Fireball make sure to disable HT of the Core i7 in the bios.
Numerous people confirm negative impact - especially in terms
of texture loading. Both XP and Win7.

FYI, I just turned off HT in BIOS and did it again twice, and had the same results +-2fps.

Also try disabling the C states and the EIDS dynamic clocking thingy in BIOS.
(so all power saving options of the CPU)
Helped me a bit.

While the Benchmark might be great per se, the only thing that made my rig stutter so badly (down to 3 FPS) was the ViewDistance setting on 3600.

As soon I set it down to 2011, my benchmark looked like in the latest attachments.

With patch 1.03 and view distance set to 6647 I got the attached results!

I reduced view distance to 2025 and did another two runs (attached).

In contrary to when I lower the view distance, the highest fps goes higher, and the average (low) frame rate stays about the same.

The interesting consequence is between my last two measurements is, that the highest framerate and the averages slightly dropped with higher view distance but the average (low) frame rate increased by 3-4 fps, while the lower view distance was be having as earlier.

I think that is what was meant to be addressed basically and its IMO a valid compromise (if that's the only solution).

Btw. with my original 3600 view distance it looks like in attached screenshots.

Fireball wrote:

Btw. with my original 3600 view distance it looks like in attached screenshots.

I.e. it's exactly the same average FPS as before, with a chance that stutters are decreased.

It seems that buildings in view cause a high cpu load, this can be easily tested by running two monitors with 1 monitor showing CPU usage.

Simply look at a forest/distant area and then look at some buildings/city. The CPU usage will increase by up to 40%, in my case maxing my Core2Duo @ 3.5ghz out.

The GPU should be under more strain in the city areas than the CPU, is the CPU doing too much work for GPU?

Yapab: I came to exactly the same conclusion as you did. In city areas, the framerate is extremely CPU limited. I can set all of the settings (including resolution) to the very lowest values and see very little improvement in framerate versus setting everything to much higher values.

Bohemia: I believe this is a problem with the way you're managing (or not managing) your API calls. DirectX performance is extremely sensitive to the number of state changes and especially draw calls each frame. My guess is that in city areas the number of state changes or draw calls is getting high enough that the overhead from the DirectX API is hogging CPU time. This is pretty well documented, a short google search turns up these two presentations from ATI and NVIDIA:

http://ati.amd.com/developer/gdc/D3DTutorial3_Pipeline_Performance.pdf
http://developer.nvidia.com/docs/IO/8230/BatchBatchBatch.pdf

I am also experiencing this issue. I hope as a community we can resolve the issue.

I have added a slightly modified version of the mission, as used in the regular autotest used here.

While I do not think I have already fixed the issue completely, build 59841 should bring a significant improvement.
Technical details: There was a bug in streaming stage 0 textures in Super shader materials, causing some file operations to be synchronous without a real reason.

Build server autotest measurement:

before: FPS: 35.6104; FPSmm 21.7391 FPSm: 3.80228 - Chernogorsk (1959 frames in 55.012 s)
fixed: FPS: 40.0122; FPSmm 22.7273 FPSm: 12.0482 - Chernogorsk (2216 frames in 55.3831 s)

Beta patch released. build 59857
http://www.arma2.com/beta-patch.php

Please confirm:

[59842] Reduced town/village stutter (fixed streaming of textures used by Super shader).

http://www.arma2.com/beta-patch.php

Haven't run the Cherno benchmark yet, but from my tests yesterday, my impression is that:

+ stutter around towns has been greatly reduced (confirmed streaming fix)
- average framerates around large towns like Chernogorsk are still very low

When moving toward towns, stutter is now almost gone. This is especially apparent when flying. From the air, performance around Chernogorsk is also fairly good. Not so much on the ground though.

Even on lowest video settings, Chernogorsk does not perform satisfactorily. On my computer (Intel Q9650, NVidia GTX285, 4GB DDR2 1066 RAM) when I put my graphics settings to very low (Res: 1920x1080, VD: 3000, AA & AF & PP disabled, rest very low) I still get maximum framerates around 40-45 in Chernogorsk, while in mainland Chernarus I can achieve >100fps.

MadDogX wrote:

Haven't run the Cherno benchmark yet, but from my tests yesterday, my impression is that:

+ stutter around towns has been greatly reduced (confirmed streaming fix) - average framerates around large towns like Chernogorsk are still very low

When moving toward towns, stutter is now almost gone. This is especially apparent when flying. From the air, performance around Chernogorsk is also fairly good. Not so much on the ground though.

Even on lowest video settings, Chernogorsk does not perform satisfactorily. On my computer (Intel Q9650, NVidia GTX285, 4GB DDR2 1066 RAM) when I put my graphics settings to very low (Res: 1920x1080, VD: 3000, AA & AF & PP disabled, rest very low) I still get maximum framerates around 40-45 in Chernogorsk, while in mainland Chernarus I can achieve >100fps.

I fully agree with MadDogX here as i have the very same observations.
Framerate direct inside the town is improved in the latest beta-patch a bit and stutter much removed, but still performance not "good" enough IMHO. Place a few Soldiers Vehicles around the City and actually try to play there is close to impossible even with the fastest Machines on the Planet (Brother1 for instance has new ATI HD5870 Card and Core I7, but still not good enough for Arma2 in a acceptable quality setting???)
I have HD4890 Toxic 1GB with latest drivers and a 3.4GHZ Quadcore Intel.

Updated by Suma 1 Tag ago

• Abgabedatum geändert von 2009-11-14 zu 2009-12-24
  24th December..... Patch 1.05 for Christmas ;-)

Place a few Soldiers Vehicles around the City and actually try to play

When reporting problems, please, always add repro, even if the repro seems obvious. Never use words like "any", "some", "a few". Either provide exact steps how to create the missions, or better, provide the mission as attachment. Instead of "actually try to play" write steps what I should do in the mission. Playing styles may be different and some may show the problem while other ones not. Should I stand on a spot and fire? Should I turn? How much? When? Should I run?

Unless done this way, you can never be sure the issue I am seeing (and solving) is really the same issue which is bothering you.

Original repro (bechmark mission) is excellent in this respect, and what I intend now is to work further on increasing the minimal fps in this mission. If and how this will help other situation is unclear and cannot be tested unless someone provides testing scenarios for them.

Yeah, the general FPS issues are likely not be related to the initial supershader issue, i.e. the stuttering.

Due to the generic title we could possibly dynamically convert this bug to suit your need, but we shouldn't; the original description says:

Everytime im turning arround getting buildings into my view framerate drops to even zero for a second or two. This makes playing at urban regions very difficult if not impossible.

That refers doubtlessly to the streaming bug, i.e. the stuttering (which is still in progress of a clean fix).

We should report "low FPS in cities" to an own bug - would you want to go ahead mr.g-c and file a new bug with a repro (video/mission)?

When reporting problems, please, always add repro, even if the repro seems obvious. Never use words like "any", "some", "a few". Either provide exact steps how to create the missions, or better, provide the mission as attachment. Instead of "actually try to play" write steps what I should do in the mission. Playing styles may be different and some may show the problem while other ones not. Should I stand on a spot and fire? Should I turn? How much? When? Should I run?
Hehe, yes of course you are right on this, i just wanted to post my overall observations of direct comparing 1.04 and latest beta 1.05 - also in a praxis way means -> actually playing with/against AI.
Furthermore the old ticket name was: " Performance breakdown in cities and even small villages" , which i thought it is/was not mainly related to the stuttering alone, which is successfully fixed now.

Original repro (bechmark mission) is excellent in this respect, and what I intend now is to work further on increasing the minimal fps in this mission. If and how this will help other situation is unclear and cannot be tested unless someone provides testing scenarios for them.
Yes i also tried the test scenario with both 1.04 and 1.05 and as i already said, i was also able to notice a very noticeable improvement! :-)
However i was just questioning the results as i don't know how much "praxis-improvements" will we get from it? Its maybe a odd example, but its maybe like when a car manufacturer claims: "Our car consumes 7L per 100KM on our test-scenario", then you buy the car and you consume 11L per 100KM ;-)

And to not get me wrong there too, i am pretty impressed overall! I thought such bugs like this one here might never get fixed - > Makes me very happy to see that i was wrong and that you are working so eagerly for a fix on this/these bugs. ;-)

That refers doubtlessly to the streaming bug, i.e. the stuttering (which is still in progress of a clean fix). We should report "low FPS in cities" to an own bug - would you want to go ahead mr.g-c and file a new bug with a repro (video/mission)?
Maybe this is indeed a good idea, but how should such a mission look like? Maybe a flyby over a 5 minutes firefight in chernogorsk also with min/average/max FPS?

mr.g-c wrote:

Maybe this is indeed a good idea, but how should such a mission look like? Maybe a flyby over a 5 minutes firefight in chernogorsk also with min/average/max FPS?

Flying over a firefight is certainly a good idea, but after all you could also prove it without firefight, to make it clearer; you could have a flight around/over chernogorsk with all the factories close up and then fly to the countryside to show the difference. 1-2 minutes would be sufficient.

I don't know how what the best way is to prove your case, but it sounds like a beginning to file a ticket and describe the issue, so we can track it - repros can then be contributed and attached by anyone.

Because soldiers put through the editor in a city was mentioned, it could also be more a problem of local AI view blocking through more building models, than just pure GPU rendering.
Two separate SP repros with and without AI could confirm this.

I have noticed an inconsistent FPS problem with the Chernarus Island "intro" scene that plays behind the main menu screen
Its covered in this thread http://forums.bistudio.com/showthread.php?t=90686.

Add -world=chernarus to the Arma2 shortcut to use the scene when the game loads.
Basically, as the screen pans from right to left my FPS drop from 60-100fps(depending on each scene), to 30 fps on each of the three scenes.
Hopefully it can be of some use for this testing as it is easy to reproduce.

CPU :Core2Quad q9300 @3.19ghz
Mobo: XFX780i
GFX: GTX295 (191.07)
Ram: 4gb of Pc8500
OS: XP64

Beta patches and non beta 1.04 all do the same thing. I'm pretty sure it has done it since release.

I have also seen this happen on a 2nd pc I have access to and with various hardware/os : Vista64, Core2Duo, 9800gtx, 2*9800GTX SLI, GTS250. (various display drivers)

UPDATE: Here is a Pic of My Taskmanger showing CPU usage while the scene is running:
http://i72.photobucket.com/albums/i192/seanysherry/ChernFPSdrops.jpg

UPDATE: This problem is directly related to View Distance. Here is what I have found: http://forums.bistudio.com/showpost.php?p=1507990&postcount=14

Here is a kind of mission template: test-camera-Chernarus.sqm. You can create a camera path with waypoints and have some stats in the arma2.RPT. It's a template, so not intended to reproduce anything using directly as it is. People that know the "hot spots", please arrange the waypoints there and add your mission+arma2.RPT here.

The ArmAIIMark might be useful too.

It is used by the community as benchmark.
It is not so much about cities, but about different situations
(harbor, airfield, open plain, 360 over woods and zoom in from the sky)

I'm pretty sure #6004 is connected to texture streaming issues, since before the rendering window dies or A2 crashes, textures begin to load really s-l-o-w-l-y.

Alot of emphasis is being placed on video cards and CPU. I was surprised that the stuttering was solved with purchasing an SSD. Other BIS forum topics even discuss using large amounts of RAM for a "RAM disk" solution.

I think the way textures are being read from the hard drive is a large part of the problem. See the following screenshot of Win7 Performance Monitor during the short ArmaIIMark test 1, and notice the frequent seek latency for PBOs during the HDD test:

http://farm3.static.flickr.com/2460/3989326374_f4330ccdc3_o.jpg

The SSD is of course much faster, which I think solves the stutter:

http://farm3.static.flickr.com/2598/3989326504_26e4c784e3_o.jpg

Using an SSD has solved all stuttering for me. Perhaps this is due to the low response time when loading PBOs in urban areas, such as Chernogorsk or Elektroz. Not everyone can be expected to buy an SSD, so perhaps a fix for preloading the PBOs and caching more textures in memory would be the fix? This is beyond my technical knowledge.

Interesting @longtime..... Maybe a solution is for BIS to make the game finally able to adress and really use more than 2GB Ram. Largeadressaware Flag? 64bit Excutable?
There is a solution for sure and in my circle of friends/relatives i not know a single person these days with less than 4GB Ram. Gamers tend to even have 6-12GB these days (i have 8GB).

Even if only the most important ones like buildings.pbo, plants.pbo, animations.pbo, etc. would be permanetly cached in RAM would Add much i think.

Well, little update, stuttering in cities and even small villages is in OA too, especially in Zargabad.

Confirmed, the performance in Zargabad in impressively lower than general Takistan.

22-35 FPS on takistan
12-19 FPS in Zargabad central town!

E6400
3G RAM
8800GT
Win7 64b

installation A2 Combined Op 1.52

Probably because OA is still more an ArmA 2 1.05 than 1.07 enginewise ?

Means all the ArmA 2 optimizations of the last time didn't make it into OA yet.

Adapted the FPS test to OA.

This is an area on the very north of Takistan where we get low FPS (15-30)
in a real mission (5on5 PvP - no AI; other missions run at 40-60 FPS on Takistan.

While in the mission here I have mostly maxed out 60 FPS,
there are four areas that seems problematic.

3) The pink trees seems to suck a lot of performance compared to the other vegetation.
4) There is a very noticeable drop at some ruins in between.
5 and 6) At the very end getting close to that one vehicle shelter it drops and
at the final camera scene around that building.

Overall FPS data:
"FPS-STAT FPS: 57.875; FPSmm 50 FPSm: 15.625 - Takistan1 (3184 frames in 55.0151 s)"

Here is another sample on Takistan in the NW town Nur.
It is a house oriented bench.

This one compared to the one above runs far worse.
Several longer phases with crashing from 60 FPS down to 15-20,
lots of stutter and even a few short freezes for a brief moment.

"FPS-STAT FPS: 50.3181; FPSmm 38.4615 FPSm: 2.89017 - Takistan2 (2768 frames in 55.01 s) "

My settings. Object detail to normal did not help in the second test either.

Q6600, 4890, 3 GB, Intel M SSD, 1 GB Ram drive. 10.5 driver. Audigy2.

OMG, what is with the Trees in your screenshots Kju??? They look really bad... there seems to be a LOD Problem aswell, isn't it?

It does not look like this while playing.

I just got to make the screenshot the moment the FPS drop appear that made
them shown in that low LOD level (or these trees caused it?).

Ingame I have never seen them look like this. Anyway there are rather severe performance
issues in some specific areas while the rest runs great.

This is the 2nd mission..

With the beta patch especially the first one has improved.
There is still a drop at the pink trees (60 to 55), and a few
very short freezes - most important at the end at the vehicle shelter
it drops to 40 with very noticeable mini freezes.

The second one also performs better; still the basic problematic areas
remains with many short or even longer freezes and drops to 20-30 FPS
for some seconds.

It looks like not a tree problem. I've tested _TEST_FPSxxx1.Takistan and I have FPS drop from 55 to 8. Something is wrong in rendering. Geometry is fine. Please check my post here
http://dev-heaven.net/issues/11739#note-4

Latest beta results:

_TEST_FPSXXX1.takistan.7z
"FPS-STAT FPS: 57.6811; FPSmm 52.6316 FPSm: 10.5263 - Takistan1 (3184 frames in 55.2 s) "
"FPS-STAT FPS: 57.875; FPSmm 50 FPSm: 15.625 - Takistan1 (3184 frames in 55.0151 s)" (old)

_TEST_FPSXXX2.takistan.7z
"FPS-STAT FPS: 55.3533; FPSmm 47.619 FPSm: 6.75676 - Takistan2 (3056 frames in 55.209 s) "
"FPS-STAT FPS: 50.3181; FPSmm 38.4615 FPSm: 2.89017 - Takistan2 (2768 frames in 55.01 s) " (old)