70256-70313 - which gives you best performance?

[suma 8]

There is a command line option in 70256 which should allow you to check if some experimental optimizations help performance on your system. Please, post your feedback here.

This should help us to decide which optimizations should be included in the "stable" patch as default.

I have updated the poll. I am sorry to have invalidated the votes of those users (they were two) who have already voted, but getting Dual/Quad core information separately can help a lot.

----

Note: this topic may be severely moderated by me to keep only the information which seems useful to me. If you are not happy with this, post your ideas in another topic.

Edited May 10, 2010 by Suma
Added guidelines.

[otrebla_snake_ita 2]

What do those number, precisely? I read

-"0 = no extra threads" (so, what the -exThread parameter does in this case?)

-"1 = file operations" (what you mean with "file operations"? Faster loading of files, like with the ramdisk?)

-"3 = texture loading" (I have no questions for this, I think XD)

Thank you :D

[KeyCat 131]

Thanks for the update...

Just did a quick test with -exThreads=0 and this seems to fix the stuttering problems I personally experienced in builds 69645 to 70184! So far I only had time for a quick test with Video Memory set at Default but it felt pretty much the same as with 1.05 final, will try other Video Memory settings as well as -exThreads=1/-exThreads=3 and report back as well as cast my vote later...

Windows XP Pro

Intel C2D E6850 @ 3.41 GHz

2 GB RAM @ 1066 MHz

2 x Seagate ST3250620AS SATA-II 250 GB HDD

GeForce GTX 260 896 MB with drivers 191.07 WHQL

Command line: "D:\Games\ArmA II\Beta\arma2.exe" -nosplash -mod=beta -profiles=d:\games\ARMAII~1\ -bepath=client1 -name=KeyCat -exThreads=0

As previous this simple test was done on Utes with me as US infantry running in the grass alongside the runway, no other units/scripts/addons running.

/KC

[Wildgoose 1]

"1 = file operations"

Would this be a good idea for dedicated server?

Or this optimisation only good for clients?

[suma 8]

Threaded file operations means file processing (including file caching) is done on the separate thread. This is sometimes faster and sometimes slower (threading always comes with an overhead).

[rundll.exe 12]

I did some quick testing with the standard benchmark missions and the results are as follows:

1.05 (no extra command line options):

Benchmark 1 16fps

benchmark 2 7 fps

1.05 70256 (no extra command line options)

benchmark 1 15 fps

benchmark 2 6 fps

1.05 70256 (-exthreads=1)

benchmark 1 14 fps, textures seem to load slower compared to 1.05

benchmark 2 6 fps

1.05 70256 (-exthreads=3)

benchmark 1 14 fps, textures seem to load faster compared to 1.05

benchmark 2 6 fps

System:

core 2 duo E4300 @1,8Ghz (OC to 2,25 Ghz)

4GB DDR2 RAM

GTX260 896MB

Windows 7 x64

This simple test shows that 1.05 gives the best fps overall, but I have the feeling that its much smoother with -exthreads=3 with this beta, however the stuttering is not gone.

Maybe the extra threads are only usefull on a quadcore or more, since other arma threads already fill my dual core cpu-time anyway?

[VictorFarbau 0]

Thanks for this update and please allow a few days to come back with feedback. I won't have the time to play Arma for hours each day but I am very willing to test all options in extensive, heavy-loaded SP and MP missions. The changes seem very important to me so due diligence is required.

VictorFarbau

[SaOk 112]

For me the all exThreads seem to work quite the same way with indentical fps (I have Q6600), but the game was already very smooth for me. The fps loss is still the same when deleting AI units. On Utes, FPS is about 50 when standing alone on map, with 13 AI groups its 28 and after deleting them ingame the fps only get back to 35. So AI gives only about 33% performance back when deleted. And 0% between alive and dead.

Edited May 10, 2010 by SaOk

[Zipper5 74]

exThreads = 0:

Benchmark 1: - 32 FPS

Benchmark 2: - 15 FPS

exThreads = 1:

Benchmark 1: - 32 FPS

Benchmark 2: - 14 FPS

exThreads = 3:

Benchmark 1: - 32 FPS

Benchmark 2: - 13 FPS

My ingame graphics settings.

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

So, as you can see, it has a rather negligible effect for me. Didn't notice an increase or decrease in stutter either. Will do some more tests and see if I can come up with anything further to report.

[h - 169]

Option 3 best here. Textures load fast like in the few previous builds and basically stutter free.

0 and 1 almost identical, both stutter quite a lot. With option 1 textures load faster than with 0 though, seemed to load pretty much the same as with option 3 (but had the stuttering).

PC specs:

Win7 Ultimate 64bit

Q6600 @ 3.2Ghz

4Gb RAM

ATI 4890 1Gb (OC'd)

Abit IP35 Pro motherboard

[suma 8]

So, as you can see, it has a rather negligible effect for me. Didn't notice an increase or decrease in stutter either. Will do some more tests and see if I can come up with anything further to report.

If you rely upon benchmarking, it would be better to use some benchmark which tracks not only average, but also minimum and median fps, as new optimizations are targeted to improve mostly the low FPS spikes.

[W0lle 1052]

Q9550 @ 2.83 GHz, 6 GB RAM and GTX280

Used the first benchmark mission and in window mode:

70184: 35 FPS

70256 no exThreads: 36 FPS

70256 exThreads=1: 35 FPS

70256 exThreads=3: 35 FPS

So no difference at all for me.

[ginger mcale 11]

CAUTION, tests was did with ArmA 2 and Windows running from a SSD, tests with a normal HDD should give another results.

Propably the best mission to test it:

_TEST_FPS6-Chernogorsk.Chernarus (Mission source from Suma from issue tracker posting #30)

exThreads .... 3 | 1. | 0

AVG ............ 38 | 36 | 39

AVG min ....... 27 | 26 | 28

highest ........ 50 | 47 | 54

lowest ......... 13 | 14 | 12

Another benches; they are with AI, slightly different things happen on every run, propably not the best missions to have exact results.

exThreads ..... 3 | 1. | 0

OPFMARK ...... 39 | 39 | 44

BENCHM. 1 .... 45 | 45 | 48

BENCHM. 2 .... 18 | 19 | 19

Computer Specs:

Intel e8200 dual core (OC to 3,8 ghz)

ATI 4850 - 512 MB RAM (OC)

4 GB RAM

Intel Postville G2 SSD 80 GB

Windows 7 - 64 bit

Windows and ArmA 2 are running from the SSD, all programms closed, unneeded services stoped, no anti-vir or other security things running.

ArmA 2 has been restarted after every mission / test run to have the same conditions.

Edited May 10, 2010 by Ginger McAle

[otrebla_snake_ita 2]

1.05 patch

Bench 1: 22fps

Bench 2: 10fps

70256 withouth any startup parameter

Bench 1: 20 (but I had a lot of stuttering)

Bench 2: 8

-exThreads=0

Bench 1: 21

Bench 2: 8

-exThreads=1

Bench 1: 20

Bench 2: 8

-exThreads=3

Bench 1: 21

Bench 2: 8

Is this possible? the same fps result with the beta? ò.ò

my startup line:

"C:\Program Files\Bohemia Interactive\ArmA 2\beta\arma2.exe" -mod=beta -nosplash -exThreads=N

Anyway, can you tell me other benchmarks which tell me minimum and maximum FPS too?

Thank you

PC:

CPU: Athlon 64 x2 6000+ (3.03ghz, minimum overclock xD)

RAM: kingston 800mhz DDR2 (1x4)

Video Card: nVidia Point of View 9600gt

Hard Disk: SATA2 Maxtor (260gb)

Mobo: asus m3a

Edited May 10, 2010 by otreblA_SNAKE_[ITA]

[Zipper5 74]

If you rely upon benchmarking, it would be better to use some benchmark which tracks not only average, but also minimum and median fps, as new optimizations are targeted to improve mostly the low FPS spikes.

Sure, no problem.

Propably the best mission to test it:

_TEST_FPS6-Chernogorsk.Chernarus (Mission source from Suma from issue tracker posting #30)

I used this test mission this time. I'm surprised you guys didn't include this as "Benchmark 3". :butbut:

Settings, specs and target line all remain the same.

Each time I tested it, I ran it at least once through to let all the textures load before recording it.

exThreads=0:

Highest: 35 FPS

Average: 26 FPS

Average minimum: 18 FPS

Lowest: 9 FPS

Comments: Was pretty smooth apart from 2 areas where the screen became cluttered with buildings.

exThreads=1:

Highest: 31 FPS

Average: 24 FPS

Average minimum: 18 FPS

Lowest: 10 FPS

Comments: Less smooth than 0, more stuttering throughout.

exThreads=3:

Highest: 32 FPS

Average: 25 FPS

Average minimum: 18 FPS

Lowest: 9 FPS

Comments: Definitely the smoothest out of the three with only 1 half-a-second stutter, despite the lower FPS.

So overall it appears that without the exThreads command I get a higher FPS, but with exThreads=3 it's a smoother experience, despite the FPS drop.

Edited May 10, 2010 by Zipper5

[RavoC 10]

I didn't do a benchmark but it loads slightly faster and feels somewhat less laggy when looking around on exThreads=3.

I don't know if you can load all exThreads or will it only make it slower..?

Anyway, when like zooming in normally I sometimes had this slow motion effect, not post process but pure lag. It's still there but a LOT less.. great improvement for me.. don't know on the battle side of things though.

Edited May 10, 2010 by RavoC

[SASrecon 0]

There's significant zoom lag in =1, but imo it was also a tiny bit more stable (I'll do some more testing when I have more time).

I'd definitely go with =3 though...

[Potatomasher 0]

-exThreads=0

Bench 1: 19

-exThreads=1

Bench 1: 18

-exThreads=3

Bench 1: 19

I played a little with - exThread=3 option on and it really felt smoother than ever. But it may very well be just my "ass feel" and in reality it performed as before. ;)

But anyway i'm just happy to know that something is done to engine code and devs are trying the best they can. Thanks BIS. :)

[knan 10]

I see very little difference between them. exThreads=1 is somewhat slower than the other alternatives. Then again, I run on SSD.

(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?)

[SnowSky 12]

I've tested it, and my benchmarks aren't really different between the versions/Optionparameters, but: The benchmark test #2 was always stutting for a very small little time when the C130 came into my view. Now with the -exThread=3 it went also good there.

As I said - the optionparameters and versions didn't change a lot for my Performance:

Version|-exThread=#|BenchMark:01[Fps]|BenchMark:02[FPS]

1.05|X|50|20

beta|0|51|20

beta|1|49|19

beta|3|50|20

The average FPS could be easily changed by some small events, so the difference of 1 Frame is not a reason because of a Patch I think. But as I said, exThread=3 maybe helped when fps is normally low.

My System:

Asus P7P55D

Intel i5 750 @2,66 (Quad Core)

2x2GB DDR III 1333 NonECC CL9 (4GB Ram)

Sapphire ATi Radeon HD 5870 (10.4 Driver)

Windows 7 Home Premium 64Bit

And at the end one thing some may think it won't be notable: Avira Antivir PE :)

[Wolfrug 0]

Hm. As with all beta patches I've seen little to no improvement, but exThreads=3 FELT a little less stuttery - the LODs loaded faster, less popping etc. I had a more uniformly crappy FPS :D But honestly, there's no real difference for me.

Anyway, the benchmark results are boring: I had an average FPS of exactly 19 in every last one of them, including 1.05 vanilla (using Benchmark 1, Benchmark 2 is like 6 FPS average for me).

System specs:

AMD Athlon 64 X2 3800+ (Dual core)

NVidia GeForce 8600 GT 512 MB

2 GB DDR RAM (the old type, not DDR2)

Yeah, I do wonder how the damned thing can even chug along still...:rolleyes:

Regards,

Wolfrug

[turin turambar 0]

exThreads=3 for me!

Mission Benchmark: Test_FPS6_Chernarus by Bohemia

Line used: -noSplash -noFilePatching -world=blank -exThreads=3 -mod=beta;mods

Mods used: Landtext mod and vopsound, apart from a few extra mods not used in the benchmarkd

Results (Avg, Avg Min, Max, Min):

exT=0

26

20

33

11

exT=1

26

20

37

11

exT=3

28

21

43

12

Computer:

Core 2 Duo E6850 3.0Ghz@3.6Ghz

4GB Ram Dual Channel

Arma 2 in 10K RPM HDD

Ati Radeon 4890

Windows 7 64bits

Apps in background: Nod32, Xfire.

edit: and in general, much less stutter than in 1.05.

[NoRailgunner 0]

Hi tested with Duo Core @3Ghz Vista 32bit - ingame settings:

viewdistance 2500m, all others on "Normal", except shadows="High" and postprocess effects="off". No addons and mods used!

general error lines:

RegisterRawInputDevices failed with 0x57

Item str_disp_server_control listed twice

Item STR_USRACT_ZEROING_DOWN listed twice

Item STR_USRACT_ZEROING_UP listed twice

RegisterRawInputDevices failed with 0x57

VON slider is still overwritten with something "Translate..." (why translate - VON = best choice)

Ok, here are the results with _TEST_FPS6-Chernogorsk.Chernarus:

-exThreads=0

"FPS-STAT FPS: 24.2407; FPSmm 16.9492 FPSm: 3.61011 - Chernogorsk (1344 frames in 55.444 s) "

-exThreads=1

"FPS-STAT FPS: 23.0914; FPSmm 17.2414 FPSm: 3.27869 - Chernogorsk (1280 frames in 55.432 s) "

-exThreads=3

"FPS-STAT FPS: 24.3606; FPSmm 18.1818 FPSm: 3.61011 - Chernogorsk (1344 frames in 55.171 s) "

and with 1.05

"FPS-STAT FPS: 25.2288; FPSmm 17.5439 FPSm: 4.04858 - Chernogorsk (1392 frames in 55.175 s) "

Test with Benchmark1: only a decrease of 1 fps from patch 1.05

Test with Benchmark2: decrease up to 6fps - best fps results with patch1.05 followed by -exThreads=0, -exThreads=3 and -exThreads=1.

Would be good to have more test and benchmark missions so people can test and report their results. Aren't such missions are better for optimizations and improvements instead of relying on "It feels...", "I guess...", "I think..." ?

[suma 8]

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. The threading optimizations are more likely to be efficient when there is more work to do (like some AI, some rendering, some sound, some file loading). Chernogorsk benchmark is definitely stressing, but its load is quite unbalanced, as no AI is present.

I know it is difficult, as benchmark numbers are a lot easier to compare, and more objective, but some more real.

The benchmark can at least serve as a kind of "bottom line", to see there are no serious problems caused by the optimization. I am not sure what better alternative could I suggest, other than: try playing and say how it feels.

[EricM 0]

It's a close call for me... so far the results in benchmarks (min/max/avg) would make it even between 1.05 and exthreads3=3.

1.05 seems to have the best peak performance, but beta exthread3 is close and the difference between max and min are smaller, hence maybe the slightly more "smooth" feeling.

To my surprise, I actually won 1 fps in benchmark 2 (the more stressfull due to AI) with the beta...

Will do some more tests with more AI...

Edit : Q6600 @2.6Ghz and SSD