Arma 2: OA: Beta Build 82282 - Memory allocator test

[Coffeecat 10]

thx for the headsup Suma!

OMAC, just copy the folder in your Arma2 directory under missions and start the mission via editor.

[NeMeSiS 11]

Are you? That is strange, as -openal still works for me (though with some quirks). :j:

You cant blame him for that, its not even documented! ;)

Out of curiosity i tried it, and it didnt seem to complain about me never installing openal on this PC before, and after installing it i still could find no difference in quality or performance with a X-fi with sounds set to 128 in benchmark 2.

[OMAC 254]

thx for the headsup Suma!

OMAC, just copy the folder in your Arma2 directory under missions and start the mission via editor.

Thank you, Coffeecat!

--------------------

I played it from my Scenario list, but couldn't figure out how to play it in the editor. There was a little stuttering at the beginning (beta 82448, not this one), but after that it ran fine. After the e01 bench was over, I found myself near good ol' Novy Sobor. I've never seen Chernarus look so beautiful. Did you enhance the colors or something, or remove some realism filter that dulls the colors for normal A2? The whole place looked like eye candy, like Crysis!

Edited July 2, 2011 by OMAC

[DBGB 10]

I have tested on a 24 GB machine and had no crashes in any malloc 'setting' - though the cpu's are only 2.7 Ghz - I forgot to turn down the eyecandy - so everything ran at 1280x720 max eyecandy and with VD@3600 - with this all mallocs hit between between 21-23 FPS. Actually malloc=4 hit the highest - 24 FPS.

Will try to turn down eye candy to see if I can get some bigger differences later.

Specs: CPU Opteron 2.7ghz GPU Radeon 5870 1 GB

Edit again: This is with 82448

Edited July 5, 2011 by DBGB

[stk2008 14]

I have tested on a 24 GB machine and had no crashes in any malloc 'setting' - though the cpu's are only 2.7 Ghz - I forgot to turn down the eyecandy - so everything ran at 1280x720 max eyecandy and with VD@3600 - with this all mallocs hit between between 21-23 FPS. Actually malloc=4 hit the highest - 24 FPS.

Will try to turn down eye candy to see if I can get some bigger differences later.

Specs: CPU Opteron 2.7ghz GPU Radeon 5870 1 GB

Edit again: This is with 82448

Nice I hope you run ARMA2 on a RAM drive if not why :P

[DBGB 10]

Nice I hope you run ARMA2 on a RAM drive if not why :P

Not really necessary the Arma2 install runs of a Pair of SSD's in RAID0 - and stuff tends to get cached after the first run.

Any way here's some more test results. It's the two chenaurus / arma 2 benchmark missions I each ran through 5 times.

It's on a computer with 24 GB and 16 cores@2.7ghz

Test results produced with these settings

1280x720 3d res 75%

VD = 555 eye candy turned to very low (except texture detail = normal; Vid-mem = high)

First run B01 / B02

malloc 0 = 54 / 16

malloc 1 = 58 / 17

malloc 2 = 59 / 16

malloc 3 = 57 / 17

malloc 4 = 55 / 17

Second run B01 / B02

malloc 0 = 53 / 16

malloc 1 = 59 / 16

malloc 2 = 58 / 18

malloc 3 = 59 / 17

malloc 4 = 57 / 17

Third run B01 / B02

malloc 0 = 57 / 17

malloc 1 = 56 / 15

malloc 2 = 58 / 17

malloc 3 = 57 / 17

malloc 4 = 58 / 15

Fourth run B01 / B02

malloc 0 = 55 / 17

malloc 1 = 59 / 18

malloc 2 = 59 / 16

malloc 3 = 55 / 19

malloc 4 = 54 / 17

Fifth run B01 / B02

malloc 0 = 54 / 18

malloc 1 = 57 / 16

malloc 2 = 58 / 17

malloc 3 = 57 / 17

malloc 4 = 57 / 17

averages B01 / B02

malloc 0 54,6 / 16,8

malloc 1 57,8 / 17,0

malloc 2 58,4 / 16,8

malloc 3 57,0 / 17,4

malloc 4 56,2 / 16,6

So the best for my configuration seems to be malloc 2 or 3.

But it also seems my B01 results are capped at 60 fps/vsync limit

I'm really looking forward to what is behind thos numbers - and what would be the optimal one for a specific computer configuration.

[woore 10]

Malloc=3 is good for me, malloc=4 is the best perfomance and NO ONE CRASH in 3 hours of playing warfare. With malloc=3 no crashes too, but with malloc=4 I have better perfomance

[maddogx 13]

I still want to know what the different mallocs are and why some of them cause crashes for some people. On my comp, malloc=4 causes the game to crash at the title screen, for example, even though Arma2 has always been rock solid for me.

Malloc=3 seems to be the ideal one, but it would be interesting to find out what's behind the setting.

[carlostex 38]

Hey DBGB which Opteron is that? Is that Barcelona? Shanghai?

[DBGB 10]

Hey DBGB which Opteron is that? Is that Barcelona? Shanghai?

It's Shanghai's 8384 in a Tyan 4985-E board. (4x)

I might be able to optimize memory 'bandwidth' / Game FPS by locking the arma2.exe to a single socket (affinity mask) - but will wait till offical patch comes along to create that launch param and experiment.

[sickboy 13]

I might be able to optimize memory 'bandwidth' / Game FPS by locking the arma2.exe to a single socket (affinity mask) - but will wait till offical patch comes along to create that launch param and experiment.

http://www.howtogeek.com/howto/windows-vista/start-an-application-assigned-to-a-specific-cpu-in-windows-vista/

[Hanzu 10]

Old news yes, but I want to put my results in too, before I delete the file I had saved them into:

1.59.82489 Benchmark 01 22 FPS

1.59.82489 -malloc=0 Benchmark 01 23 FPS

1.59.82489 -malloc=1 Benchmark 01 22 FPS

1.59.82489 -malloc=2 Benchmark 01 20 FPS

1.59.82489 -malloc=3 Benchmark 01 23 FPS

1.59.82489 -malloc=4 Benchmark 01 22 FPS

1.59.82489 -malloc=5 Benchmark 01 22 FPS

Visibility set to maximum of 10000 which explains quite low average FPS compared to the hardware power.

[gc8 977]

Hi i have question.. why do you need us to test memory allocation? i think memory allocation should be designed to be be stable so im confused why the code needs to be tested in the first place..??

[nuxil 2]

.... i think memory allocation should be designed to be be stable so im confused why the code needs to be tested in the first place..??

did you think before writing this ??

it needs to be tested so it can be verifyed that its stable. maybe you forgot why beta testing is done

[Dwarden 1125]

it's already obsolete question, 'stable' memalloc was chosen long time ago

[gc8 977]

did you think before writing this ??

it needs to be tested so it can be verifyed that its stable. maybe you forgot why beta testing is done

have you ever programmed and done memory allocation stuff? ofcourse everything needs to be tested. just that 5 aproaches gives me funny ideas.

which i wont goto any futher since they are probably not interested to explain....

[suma 8]

have you ever programmed and done memory allocation stuff?

We have used several memory allocators written by experts, including TCMalloc, NedMalloc and jemalloc. Most of them broke down under the ArmA 2 load, causing memory corruption or crashes.

[maddogx 13]

We have used several memory allocators written by experts, including TCMalloc, NedMalloc and jemalloc. Most of them broke down under the ArmA 2 load, causing memory corruption or crashes.

So which one do you use now? A "roll-your-own" solution?

[suma 8]

So which one do you use now? A "roll-your-own" solution?

I am sorry, this is something I cannot answer yet.

[maddogx 13]

I am sorry, this is something I cannot answer yet.

That's too bad. But I guess it's not really important, as long as the game works. :)

[BasileyOne 10]

So which one do you use now? A "roll-your-own" solution?

obviously its intel-based ones.

ironically this[along with using ICC(heavily forked/tuned/tripwired GCC fork)]probably cause of such "random" crashes, with 3-rd party stuff, like memaloc and etc.

intel make quite good chips sometimes[and sometimes not so good], but monopoly mean greed and taste for lockout, IMO.

[maddogx 13]

obviously its intel-based ones.

ironically this[along with using ICC(heavily forked/tuned/tripwired GCC fork)]probably cause of such "random" crashes, with 3-rd party stuff, like memaloc and etc.

intel make quite good chips sometimes[and sometimes not so good], but monopoly mean greed and taste for lockout, IMO.

Did you even notice that the question to which you're replying is >2 months old and was answered looong ago? :rolleyes:

[Dwarden 1125]

obsolete thread, closing to prevent it constantly come up (e.g. has params in first post which aren't used anymore)