70256-70313 - which gives you best performance?

[NoRailgunner 0]

OK, I was only thinking about a proven testmission base where you can go build and rely upon. :)

Which default A2 missions can be considered as "benchmark incl. AI" or close to? What about the 1st, the 3rd and the 5th SP mission?

[mrn 0]

Seeing the results so far, I will try to direct the feedback a little bit. While benchmark scores are interesting, real life experiences from how the game plays are definitely more important.

Just after I ran some benchmarks!

Anyhoo using this:

http://dev-heaven.net/issues/1745#note-30

Ran 3 times and used the last result.

Avg/Avg Low/High/Low

1.05

22/13/38/4

60826

22/13/32/4

Exthreads=0

21/11/39/4

Exthreads=1

20/10/31/2

Exthreads=3

20/12/37/1

The key here is the lowest figure, on the later patches my game will just freeze for a split-second then carry on. This never used to happen and is unpredictable.

Same results just having a quick play in the editor 70256 "felt" a bit stuttery on Exthreads=0 and anything else was unplayable due to the freezing.

Resolution:1920x1200

Viewdistance=2393

Texture detail = High

Video Memory=Very High

Ansio Filtering=Very High

AA=Disabled

Terrain =Very Low

Objects = high

Shadow = high

PP = Disabled

Hope that helps

[mrcash2009 0]

@SUMA:

I reported since this new threading engine change was introduced since patch 69645 that I had realy awful performance, this stayed from 69645 up to and including this patch, the details I posted HERE.

So again with this patch I had the same very poor performance and nothing had changed, until I used -exThreads=3. This alone removed all my issues and brought the performance right back to what im used to since 1.05 and also beta's prior to 69645.

So to conclude, on my rig (dual core specs in my sig) what has been changed does realy effect things rather obvious unless setting this threading to whats needed and this did the trick, here's a quick video showing me on an unpopulated UTES map and the difference:

Bear in mind im spinning on the spot the same turn speed in both clips, and also its UTES unpopulated, its clearly obvious when you see the second section with the thread option added (I had the FPS counter on in Fraps but it didn't get recorded so I manually noted it).

FnH2c98F23c

Without knowing this thread option It would have been unplayable for me, so its crucial to get that right.

Edited May 10, 2010 by mrcash2009

[Richey79 10]

-exThreads=3 considerably better for me.

I don't benchmark, as I've seen the game run completely smooth one time and stuttering badly another: both times Fraps said 25/30 FPS.

Anyway, with -exThreads=3 I could run textures on very high instead of medium smoothly (I run quite a low view distance anyway). Nice and smooth gameplay experience.

I've felt for a long time that it was texture loading that was one of the two main culprits for any stutters my system encountered. The other is CPU strain caused by AI units, which is a necessary evil.

Quad core running CrossFired cards.

Thanks for the continued work on the engine, guys!

[Franklin 0]

I installed the Beta and now I cant run the game at all. I get an error saying "Shaders not valid (mismatch of exe and dta?)" My game gives me this error if I try to run with the beta or not. I ran the game with no issues minutes prior, then installed and ran the beta and I get this.

My guess is possibly an issue with the drivers related to Nvidia 3D Vision [197.45(8.17.11.9745)]

system specs:

Asus P6T6 WS Revolution NB:Intel X58 SB: Intel ICH10R

Intel® Core i7 CPU 930 @ 2.80Ghz model BX80601930

3x GeForce GTX 260 (BFGEGTX260MC896OC2E) 896MB in 3 way SLI driver 197.45(8.17.11.9745)

Onboard audio

Windows Vista 64 Service pack 2

HITACHI Deskstar 7K1000 model HDS721075KLA330 (0A35154)

[malick 0]

Definitely an improvement over latest builds wit exthreads=3 !

I don't "feel" stuttering anymore and can play with healthy 25-30 fps on Chernarus island. On Utes, fps can get up to 40-50 depending on location.

From 1.05, I won a couple fps, which is good, but the game became clearly more stable and I have a lot less LOD/texture loading issues.

Thanks Suma !

Malick

[Protegimus 0]

Suma, thanks for taking the time to explore this in conjunction with the beta tests, your effort really is appreciated!

To me, the additional threading with -exThreads=3 stands out in the very busy benchmark 2 - better audio rendering (more simultaneous audio effects) and smoother animation are the result. Comment regarding 'it's no free lunch' noted.

Need more time to give a fair report about in game, but initially the 'feel' is great.

Also appears that ArmA is making more use of the internal file cache (with -maxMem=2047 parameter) than with release 1.05.

Now it is clear what is possible with file and texture loading, can I respectfully ask if there are other options for decreasing the workload of the main thread?

Is it all rendering setup and physics stuff or are there other possibilities like an option for using GPU for all AA instead of CPU?

I see very little difference between them. exThreads=1 is somewhat slower than the other alternatives ...

(What really reduces min avg fps spikes for me is cpucount=(my number of real cores), to avoid running on hyperthread siblings... could that please be set as the default?)

No thanks knan - bear in mind that not everyone uses the same system as you and we don't want our game borked because of some quirk of a particular O/S.

My suggestion is to dual boot to 32bit WinXP for ArmA 2, then you will not require -cpuCount= and you will get better FPS for free!

...

System specs:

Windows 7 Ultimate 32bit

Intel Core 2 Duo E8400 3.00GHz

nVIDIA GeForce GTX 280 1GB

2GB DDR2 RAM

Target line:

"W:\Program Files\Bohemia Interactive\ArmA 2\beta\arma2.exe" -nosplash -cpuCount=2 -maxmem=2048 -noPause -showScriptErrors -exThreads=0/1/3 -mod=beta

...

Zipper5, note you have -maxmem=2048 parameter on your target line.

-maxMem=2047 is the limit for setting the file cache and with only 2GB physical RAM in the system the practical limit for -maxMem= is going to be a lot lower. Maybe check your free physical RAM counter to see what is available to allocate to ArmA 2, adjust and test again - or does it already give you the best results?

Protegimus

[celery 8]

There was no suitable option in the poll, so 70256 with -exThreads=3 option is the best so far, I have triple-core.

[frankyman 10]

Mission : _TEST_FPS6-Chernogorsk.Chernarus

System: Intel Quad Q9550 2,83Ghz

Geforce Gtx+ 9800

4 GB 1066 ram

Sata 2 hdd

Win XP 32

Minimum settings = Tex=low , Asin_Filter=off , AA=off , Obj=v-low , Shadow=off , post=off, 3d_Render 100% of 1600/1000

Maximum settings = Tex=v-hi, Asin_Filter=norm , AA=norm , Obj=v-HI , Shadow=v-HI , post=V-HI, 3d_Render 125% of 1600/1000

Minimum settings:

exThreads=3

av --- 59

av(min)-- 41

Hi --- 70

Lo --- 12

3296 frames in 55sec

Maximum settings:

exThreads=3

av --- 19

av(min)-- 13

Hi --- 28

Lo --- 2

1072 frames in 55sec

Minimum settings:

exThreads=1

av --- 58

av(min)-- 39

Hi --- 69

Lo --- 7

3296 frames in 55sec

Maximum settings:

exThreads=1

av --- 20

av(min)-- 14

Hi --- 29

Lo --- 7

1072 frames in 55sec

Minimum settings:

exThreads=0

av --- 59

av(min)-- 43

Hi --- 69

Lo --- 13

3248 frames in 55sec

Maximum settings:

exThreads=0

av --- 21

av(min)-- 14

Hi --- 29

Lo --- 2

1104 frames in 55sec

[Fireball 16]

I've seen Suma's remark after I've ran all the benches and I'll thus only summarize the results:

ArmAIIMark

It ran on all options pretty much the same:

Test1: 28-31

Test2: 34-36 (Hangar fly-by)

Test3: 30-31

Test4: 42-43

Test5: 15-41 (Space capsule)

So, as we can see, the differences in value are neglectable, the raw numbers are not so important here thus, except for two special mentions:

Test2, the hangar fly-by stutters considerably with 1.05 and beta (noparam) but with exThreads 1 and 3 it becomes smooth.

OTOH, Test5, the "space capsule drop", is the most remarkable of them:

1.05: 15 fps

beta: 41 fps (no param)

beta: 27 fps (exThreads=1)

beta: 31 fps (exThreads=3)

So interesting is that this is the only test which performs worse FPS-wise on the beta, but with exThreads=3 it feels definitely a lot smoother.

ArmAIIMark values:

1.05: 3131

beta (0): 3649

beta (1): 3310

beta (3): 3456

Chernogorsk Benchmark

Interesting enough I couldn't find significant differences between any version and param setting, also personally I only vaguely noticed increased stutter on raw 1.05 there.

I'll do some gameplay testing, as wished, during the next few days and report back.

Edited May 11, 2010 by Fireball

[MavericK96 0]

So from what I can gather from summarizing the thread so far, running the beta with exThreads=3 seems to give only slightly worse FPS than 1.05, but seems a lot smoother in general? Has anyone tested the new beta against 70100/70184?

Zipper5, note you have -maxmem=2048 parameter on your target line.

-maxMem=2047 is the limit for setting the file cache and with only 2GB physical RAM in the system the practical limit for -maxMem= is going to be a lot lower. Maybe check your free physical RAM counter to see what is available to allocate to ArmA 2, adjust and test again - or does it already give you the best results?

Protegimus

Is maxMem=2048/2047 a parameter that is necessary/useful? I've tried several times using it versus not using it and never saw any difference either way.

[AnimalMother92 10]

2.8ghz Intel Xeon Quad Core, 8gbs RAM, GTX 285 1gb, Intel SSD, Windows 7 x64

1.05

exThreads=0

exThreads=1

exThreads=3

exThreads=3 seems to be the best to me. Will continue to test.

[Protegimus 0]

MavericK96: best explanation of -maxMem= that I know of is here, ref.: "The Barrier Is Broken" section.

I always found texture loading was better with a larger internal file cache, no doubt everyone's mileage will vary according to their system.

Have you tried your flyby test yet with -exThreads=3?

For me, performance when flying an attack jet over Chernogorsk, especially when zooming view for targetting with the cannon gives a noticeable improvement over build 70184.

Protegimus

[MavericK96 0]

MavericK96: best explanation of -maxMem= that I know of is here, ref.: "The Barrier Is Broken" section.

I always found texture loading was better with a larger internal file cache, no doubt everyone's mileage will vary according to their system.

Have you tried your flyby test yet with -exThreads=3?

For me, performance when flying an attack jet over Chernogorsk, especially when zooming view for targetting with the cannon gives a noticeable improvement over build 70184.

Protegimus

I haven't gotten a chance to yet, due to being away from my gaming rig at the moment. Probably won't get to test until Wednesday night or Thursday morning, but then you can bet I will do some thorough testing. :cool:

I think a more interesting comparison would be this new beta with exThreads=3 versus 70100, as opposed to 70184. In my testing, I pretty much threw out 70184 because it gave me a LOT more stutter than 70100, and in my flyby test the game crashed after only 2-3 passes. I surpassed a dozen passes using 70100 with no issues whatsoever, and much less stutter than both 70184 and 1.05. The only crash I've ever experienced in 70100 was using ACE + other mods playing online, and it took almost 2 hours or so to crash even then.

Thanks for the explanation link on maxMem. I do understand what the command *does*, I'm just not certain I've ever seen a difference in FPS or stutter whilst using it in ArmA2. I guess that was what I was most interested in, seeing as my ArmA2 will use up to around 1.8 GB of RAM by itself with no maxMem parameter, and I was curious as to what yourself and other users have gotten in improvements when using the command, in addition to what sort of PC specs it's most useful for.

[Rexxenexx 0]

ArmA2Mark Saves.txt:

  -ArmAIIMark Benchmark Tests-

All settings Very High except PP=Low Shadows=Normal Res=1920X1200
[in order ran|spaces indicate application shutdown]

v1.05 = 1508

beta build70100 = 1981

beta build70184 = 1917

beta build70256 = 2003 (-exThreads not used in shortcut)

beta build70256 = 2030 (-exThreads=1)

beta build70256 = 2040 (-exThreads=3)

beta build70256 = 2205 (-exThreads=0)
beta build70256 = 2090 (-exThreads=0)

beta build70256 = 2037 (-exThreads=3)
beta build70256 = 2062 (-exThreads=3)

ArmAIIMark isn't really a perfect test. But my eyes told me -exThreads=3 was better. Less studder-loading of textures and LODs. The -exThreads=0 seemed to load textures way too late and held low poly LODs longer.

[Brainbug 10]

I think this was definitely an improvement (I'm running it with exThreads=3), looking around quickly is smoother and "snappier". Not by much because my sys could handle the game quite well already, but still it's a noticeable change to the better. It's still not totally smooth like Quake3Arena or something like that, but it's getting there. Good work, BIS!

my system:

AMD Phenom II X4 940

4 GB DDR2 RAM

Radeon 5850

SuperTalent Ultradrive ME 128GB (SSD)

Win XP 32bit

Game settings are at max except postprocessing which is disabled, AA is at 4x, 3D render res is 2400bySomewhat (i.e. 1920x1200@125%).

[CarlGustaffa 4]

Would "-exThreads=1 option, I have DualCore" be the most sensible option for a dual core server? Or is there enough texture stuff going on even on server to warrant -exThreads=3? Server runs Win7 64 bit btw.

Downloading update now.

[nerdwing 13]

Either 1 or 3 make a substantial difference I think. Overall 3 in my case, it appears.

Geforce 7950 GT

AMD Athlon X2 Processor 3860

2 gigs ram

All settings low or very low where possible, shadows disabled, AA and post processing disabled, view distance 1200.

FPS in the default benchmarks hasnt changed noticeably, but ingame feels alot more fluid during missions and campaign. Most noticeably, when entering Manhattan during the campaign I noticed very little microstutter with exThreads=3, and smoother gameplay overall it seems.

Also possible coincidence, but GPU temp is MUCH lower than it formerly was (upwards of 90 Celsius, now 80 at max), and is a very welcome side effect of the betas :)

[WillaCHilla 10]

Testet the new parameters yesterday. -There are no significant FPS-differences between them in my case.

-But for smoothness, its the "-exThreads=3" which I prefer.

Ingame-Settings and hardware in my sig.

[twisted 127]

-exThreads=3 SEEMS TO MAKE THINgs smoother and textures loads a little faster. i also use cpucount = 4 is that needed here?

[idfhayal 10]

The game was loaded with a lot of mods (ace,gl4,JTD etc.).I've checked a lot of times but I always get almost the same result.

Benchmark 1:

1.5 15-53-37 (min-max-avg)

beta (0) 22-63-41 !!!!!

beta (1) 18-63-37

beta (3) 18-63-38

Intel Core2 Quad Q6600 2.4GHz

4GB Ram

8800GT 512 MB

Windows 7 x32

[VictorFarbau 0]

For me exthread=3 is definitely an improvement. Playing a heavy mission on my local PC (up to 700 soldiers, 30 trucks, 10 Tanks, 5 Planes, 5 choppers and each unit running several scripts) results in smooth gameplay if I use setting "3". If I use "1" it feels a lot more sluggish somehow.

Odd: whatever parameter I start Arma II with, the effects won't show in the Windows Task Manager or in FPS displays - they always report about the same values (25-30 FPS, up to 48% CPU on a quadcore, ARMA II using 15-16 threads). Nonetheless, I experience the best performance when using "3".

When I host this mission on a dedicated server (which is the purpose) then my local PC runs at 55-65fps consistently, no matter if I use -exthreads or not. Seems to me that the effects become more visible the lower the PC performs to begin with.

Regards,

VictorFarbau

[pauliesss 2]

For me exthread=3 is NOT an improvement and bring stuttering to me. The best option for me, personally is without this command line.

[Headmunky 0]

My PC:

CPU:

Intel Core 2 Duo E6600 @ 2800 Mhz

Mainboard:

Asus P5B-Deluxe

Memory:

G.Skill DDR2 2Gb @ 300 Mhz

Video card:

GeForce 9800 GT @ Nvidia 196.21

ArmA 2 Settings:

Everything @ "LOW" or "VERY LOW" @ 1680x1050

My result:

Final 1.05 is best for me.

I never had any bad issues with 1.05 by the way

http://headmunky.de/FPSTEST.rar => screenshots

note: I tested without any AI. Just plain scenery

[RavoC 10]

Well I playedsome Warfare online yesterday . I don't know but I had this major lag coming into Chernagorsk.

I was in an mi-17 flying around, I saw no textures at all, ground textures far out, some I did see but it where like squares that where gone and suddenly my fps began to drop to 1 even 0 just freezed so hard it took me 10 minutes to get into my menu and get my mouse on ABORT, not even mentioning the time it took me to click so I ended the process...

This was with exThreads=3

:j: