perf_prof_branch

[ashen helm]

sadge

[whole cloud]

Well.. Most.
YAAB runs fine 😄

The lambs issue has been identified.
But v8 is affected by the lambs issue too!

[heavy galleon]

still time for v10

[ashen helm]

https://cdn.discordapp.com/attachments/105466848958513152/1210179960015159427/cooltext452791206564828.gif?ex=6735d8ea&is=6734876a&hm=d10b3dca6f0a40895bed218519163a463cf70ef01dc3050b1fc6e4282321c3e6&

[whole cloud]

https://media.discordapp.net/attachments/105466848958513152/1210179960015159427/cooltext452791206564828.gif

That is the original

oi why is the original not working

https://media.discordapp.net/attachments/
https://cdn.discordapp.com/attachments/
😠

[vivid rune]

Yeah posting pictures was violated by members of BI so it is forbidden...

[whole cloud]

Okey thanks to the dumps, I found and fixed the lambs and flashlights issues.
Bit too late for today, but in time for tomorrow morning

[whole cloud]

Bah I forgot to fix the bug in the non-chrome COPY ALL button thing. v10 for that one then :harold:

[vivid rune]

I played the Eastwind campaign. In "Signal lost" (Kerry wakes up on the Beach of Altis) the stand up animation freezes if he put his left hand in the sand for a couple of seconds. After that it cut after some seconds to the next animation. Verified it with stable. There is all fine. Dev version has the same flaw (14.11.24)

[minor geode]

So has the current build on steam addressed the blinking unit bug? I had reverted to a previous 2.18 build waiting for it to be addressed, no rush, just wondering.

[knotty wraith]

v9 is no longer available

but the flickering is fixed

[knotty wraith]

I saw the flicker again

[vivid rune]

You playing the v8 again. It is a roll back.

[knotty wraith]

I managed to download v9

[vivid rune]

Which build number are you playing right now?

[knotty wraith]

arma3client_x64_profiling_218_152408_v09

[acoustic lynx]

Has there been a change with how mission objects are created in MP vs SP on profiling? I'm noticing some vehicles and objects get placed in SP but not MP when testing in the editor on profiling.

Seems to not be just a profiling issue :\, I'll see if I can narrow down the issue and post about it in the most apt place

[primal silo]

Curious if there is a specific recommendation for a mimalloc or memory allocator I should be using other than default?

[ashen helm]

not really

if you have a lot of ram you could try using goldjohnking's mimalloc

[heavy galleon]

I am running with these, but I wouldn't expect any big differences https://github.com/GoldJohnKing/mimalloc/releases

[fallen furnace]

Hi, I'm sorry if this has been asked before; I know generally the average FPS is supposed to be better but should we also expect better 1% lows? General performance has always been decent, but in Antistasi for example FPS can be conistently low late game w/ many stutters.

[woven loom]

I did see a notable improvement with this (lock_pages variant)

[woven loom]

Because of performance binary or the malloc?

[void badger]

Considering the optimizations done, certainly

[plain trout]

You can’t make any predictions on performance. The workload is also very different for different scenarios in arma. Also most things in performance branch are still WIP. So once this crowder finds an issue with a feature, it might disappear and it usually comes back a later time. Sometimes features only get re-added in reduced or changed functionality. Some things that get optimized are for ver special uses. Until the next arma version is released consider nothing final.

[eternal kraken]

with mimalloc_v214_lock_pages and the last profiling version my YAAB avg. FPS increases from 47 to 57 🙂

https://i.ibb.co/2vhMXdQ/image.png

[spiral pond]

anyone fancy to upgrade it with storing the fps data points to MissionProfileNamespace and show the graph for the selected run?

(+ native support of console via desc.ext settings and logging slow frames if ran with profiling build)

[woven loom]

Aye I saw a good 10 fps increase too iirc

[primal silo]

Appreciate your help, this was very helpful and honestly gave me a solid avg 15 fps boost

[heavy galleon]

happy to

[primal silo]

Same to you, thanks as it really did help with performance

[whole cloud]

2.18.152418 new PROFILING branch with PERFORMANCE binaries, v9, server and client, windows 64-bit, linux server 64-bit
- Added: Multithreaded PhysX simulation
- Fixed: Crash when AI group is deleted while being simulated
- Fixed: AI visibility checks did not consider per-frame time limits
- Fixed: Crash related to weapon flashlights/IR lasers
- Fixed: Game would try to re-download Steam Workshop Compositions repeatedly - https://feedback.bistudio.com/T171116
- Fixed: Zeus map did not render "Empty" markers - https://feedback.bistudio.com/T185279
- Fixed: Difficulty could get stuck in "Regular" after a campaign intro mission
- Fixed: DataLink config attributes in "Man" type classes were ignored - https://feedback.bistudio.com/T186290
- Fixed: inArea/Array/Indexes did not support legacy hexagon
- Fixed: RPT file header would not print completely if too many mods are loaded
- Fixed: RPT file header would not print all arguments (if arguments are too long)
- Removed: One rendering optimization (caused objects to visibly flicker, pending further investigation)

Note: Profiling branch contains experimental changes that can be unstable. Please report issues to developers, and consider switching back to stable / default branch if the issues are too severe.

If you don't want to use the Steam branch, the files are also available for alternative download here:
https://drive.google.com/drive/folders/15p9j7C2nHUt6NoVfChX4YFuqzFXzblJh

[eternal kraken]

what a nice new arma feeling with those improvements. thank you dedmen 🙂 https://i.ibb.co/TbBXpBy/20241114110113-1.jpg

[plain trout]

• Fixed: Game would try to re-download Steam Workshop Compositions repeatedly - https://feedback.bistudio.com/T171116

just realised you get really bad fps (22-25) in the menu if you run with -noworld startup paramenter and 40k view distance. On prof and vanilla. Worse than on an actual map.

Also the background shadow calculations that were supposed to be fixed with 2.18 are still only deactivated by lower shadow drawing distance. Do you want FT tickets?

[heavy galleon]

I might have an unorthodox solution. Don't use 40k view distance

[inland dew]

looks like you have a fps lock, thus wrong numbers in yaab

[eternal kraken]

this is called VSync and its intended

[inland dew]

but results in wrong fps numbers in the benchmark

just so you know
not that i care

[eternal kraken]

the parts in game where i get 100 fps i dont care. but the parts where i get less then 60 fps i care a lot so

for me everything is benching as intended

[inland dew]

it's not, but ok

[heavy galleon]

Doesn't vsync make it even worse for situations where you're pushing less frames than your monitor can do?

[plain trout]

#hardware_vs_arma

[empty goblet]

it depends what type of buffering is active and arma has alternative vsync and there is also option for tripleBuffering=1; for client-side arma3.cfg

[ashen helm]

"alternative vsync"?

[empty goblet]

there are multiple types of Vsync and multiple types of buffering

the usual tearing-prone vsync is so called double-buffered

[ashen helm]

oh I am aware of this and set vsync through nvidia control panel, thought there is a setting for vsync type within arma

[empty goblet]

i don't think the tripleBuffering is definable in UI nor launcher ... it's less known feature added on my request sometime in A2 development (so works in A2. A2OA, TKOH, old dayz (not the new renderer) and A3

if anyone want to test the 2.18.152418 build on servers
Warlords XL (so with AI) 85.190.148.53:2102
KOTH (no AI) 85.190.148.52:2102
Warlords Redux 85.190.148.62:2302

just note yesterday there was like 60 players on the KOTH server with 152408, FPS in range 400 to 500, when 40 players it was over 700 😁

[hybrid cradle]

Is there any chance of getting a frame limit graphics option btw? Want to clamp lower than 144 with vsync lol

[empty goblet]

there is limitFPS -limitFPS= but that is just for no-UI server-side only (ie shall work with headless client too)

tho for frame-rate limiting i suggest to use MSI afterburner

[hybrid cradle]

Yeah I meant in the video settings bit

Would be more convenient imo

I'd figure it's most likely not in because it's just meddling with gui in rv4 which is Hate

[whole cloud]

there is a frame limit cheeatcode

[gusty viper]

Does that also work on client? Thought it was server only.

[whole cloud]

⚠️ One of the crash fixes from yesterday didn't work.
We'll replace this build by a hotfix on 152413 in about 30 minutes

[empty goblet]

not sure, @whole cloud is that server-side only or affects client too ?

[whole cloud]

limitFPS is called "FPSLimitServer" internally

[empty goblet]

i claim ignorance on this because i always use MSI afterburner and it's RivaTunerStatisticsServer to define maxFPS for each game / situation i need

[empty goblet]

ok , that answers it 😁

[whole cloud]

It also works on client, but only when rendering is disabled, so when the game is minimized it probably would do it

[empty goblet]

clientside crash or recommended server update too ?

[whole cloud]

both

[plain trout]

Performance enhancements are crazy in modded koth.

[knotty wraith]

wow, + 6 fps on YAAB (mimalloc_v214_lock_pages)

https://s.iimg.su/s/14/sdVXlCJln6DOdyyf0v8i2ulkLQopec7qOcZZHclR.png

[drifting swan]

does it make sense to install custom allocators on the server?

[whole cloud]

If the allocator is better/faster, then yes

[empty goblet]

so both servers on 2.18.152413

[whole cloud]

Fix is up now. So should now be crash safe? 🤞
Also need info about the flickering, it should be gone now

[drifting swan]

flickering is gone for me

[hallow lantern]

Still crash on linux dedi

[whole cloud]

verified its running the 413 build?
analyzing linux crashes is such a chore

[short hollow]

Crash safe including the LAMBS issue?

[whole cloud]

should be yes

[hallow lantern]

2.18.152413 yes, I will try to remove lambs and try again

[hallow lantern]

Without lambs no crash

[empty goblet]

Redux is now on 152413 build too ...

[hallow lantern]

It's the 3,5 Gb file named "core" that has the crash dump in linux, right?

[knotty wraith]

Do you have a HC client in some north?

[empty goblet]

nope, no need as no gains for any of those scenarios on A3 official servers

[knotty wraith]

very sad, I feel like I’m the only one suffering from the fact that the NS doesn’t work as it should, but I can’t provide complete information

[empty goblet]

what was the problem again ? today's build so far seems relatively stable so maybe time to make good repro/report

[empty goblet]

except this one

[knotty wraith]

I don’t know what exactly the problem is, but with v5 the HC does not work as before... I’ll just test today’s prof...

[empty goblet]

'doesn't work as before' is very non-descriptive 🤣 also 152307 is last 'stable' build before massive dump of MT tweaks, so you would need to say which exactly was the last build HC was working as before

[whole cloud]

yes. But to analyze it I also need the libraries that your server loaded into the Arma process

[hallow lantern]

I'm using this docker https://github.com/fugasjunior/arma-server-manager/

[knotty wraith]

All my players have already run away from these tests due to constant problems with the HC and due to problems in the entire game and its results, I would be glad to roll back to V5, but I have a feeling that everything will remain the same, you will be fine because you are not using HC

[empty goblet]

1st need to know what was last build which was working fine ? v4 or older ? and again what doesn't work ...

need clearly defined what is wrong , hard to fix something something unclear

is there Arma 3 Feedback Tracker ticket for this issue ? if not, make one

[knotty wraith]

If I don't do it, the problem won't go away, will it?
It’s strange to me that I’m the only one who doesn’t know English, I have to do things that carry responsibility

[empty goblet]

well crap, WL redux server just crashed

[knotty wraith]

Then I need an adequate translator from your side)

[empty goblet]

you just need to explain the problem and ideally make the ticket, just saying 'stuff is broken' isn't much helpful unless it's broken for everyone

[knotty wraith]

I sometimes report problems Dedmen, but according to his rhetoric, for my implementations I need to purchase a nuclear collider after his innovations 🤪

[heavy galleon]

Where Linux allocators?

[empty goblet]

linux allocator is hardlinked

[knotty wraith]

I just very much doubt that I, with my hardware with A2 on the server and ignorance of the English language, should here defend the HC for the right to live (that is, a performance check) - I think this is the work of the developers

[empty goblet]

dump is in dedmen's hands now 😁

[empty goblet]

again, you arguing about HC, describe the problem, make ticket ... which will allow to replicate it

in short, many lines later, i still have no idea what's wrong with Arma 3 HC except you saying so

[hybrid cradle]

Just to check - prof changes aren't being merged into main with cdlc release right?

Assuming cdlc release still requires stable update ofc

[empty goblet]

while i wish we could smuggle performance update like that, due to risk of breaking stuff, it's better to be separate updates

[hybrid cradle]

Yeah figured as much

[fickle geyser]

They're not merged.

[empty goblet]

so bad news, repeated crashes on Warlords Redux server, reverting that one to 152307

2024/11/14, 16:30:25  ➥ Context:     [] L71 (A3\Functions_F_Jets\Functions\Ejection\fn_PlaneEjectionFX.sqf)

2024/11/14, 16:30:25 Ref to nonnetwork object 1841484: <no shape> #particlesource
2024/11/14, 16:30:25  ➥ Context:     [] L76 (A3\Functions_F_Jets\Functions\Ejection\fn_PlaneEjectionFX.sqf)``` let me guess ... airplane + ejection == server crash

[whole cloud]

^ All the crashes are inside mimalloc

[gusty viper]

Should the advice here not be that people stop using these custom allocators?

They are just another noisy variable

[whole cloud]

Usually they don't crash

[eternal kraken]

but those are only server crashes not client or am i wrong?

[vivid rune]

This still happens:
#perf_prof_branch message

[empty goblet]

so what i did was check if Microsoft released new allocator version and i see they did, 2.17 is last (A3 one is at 2.14)

[naive osprey]

I did have 2 crashes earlier on YAAB with tbb4 (pretty sure its default) with v09 - 152413
Ran YAAB 4 times with each allocator (tbb4, JEMalloc, mimalloc, mimalloc lock pages) just to check for FPS differences, only crashed the twice and only with tbb4 - Could just be a fluke, or something on my end, but worth reporting.

https://www.mediafire.com/file/dav8a90hronjgdm/Arma3_x64_v09_crashes.rar/file

[restive pilot]

A3 mimalloc by @tawdry gazelle runs an async thread that periodically "frees" memory (it calls the mi_collect fnc iirc). I wonder if that's why

As in mimalloc collection is not thread safe

[fast hornet]

he's usually pretty fast to release a new version.. but it takes 1-2 weeks until battleye whitelists the new one... guess you guys could potentially speed that up 😬

[heavy galleon]

I did not encounter any crashes client side and I am running locking pages version

[whole cloud]

Is that vanilla? no mods?

[naive osprey]

Correct, no mods loaded

Can see if I can find a linked RPT for the crashes

[naive osprey]

https://www.mediafire.com/file/gi25am304z9fvul/Arma3_x64_2024-11-14_07-05-19.zip/file
Has the bidmp and the rpt for the first crash, second crash seems to have either not generated or been auto-deleted?

[whole cloud]

oki I found something wrong, there's a use-after-free atleast

I didn't test run with system allocator. tbb allocator doesn't free memory when game tells it to free, so it doesn't detect use-after-free

[empty goblet]

i didn't see GJK for half year so i hope he is fine ...

[empty goblet]

i run 2 other servers with the same allocator and 413 and those didn't crash yet (different scenarios)

[spiral pond]

no crashes here yet - running WLs on SPE. only AI in vehicles is having a hard time to advance to next sector.
most likely unrelated to the branch, but rather general pathfinding problems due to bocage landscape and object density

[restive pilot]

locked pages never decommit iirc

[empty goblet]

ye but the server which crashed has 64GB ram and half was free

[restive pilot]

I mean maybe the crash Dedmen got was access violation. i.e. page was decommitted

[empty goblet]

but if page is locked then it shall not be

[restive pilot]

Oh you have locked pages on?

[empty goblet]

ye

[fast hornet]

jumped into a plane_fighter_04 a few times and ejected btw.. no crash. Client and server 2.18.152413 .. with the locked pages mimalloc (and huge/locked pages enabled in the windows policies ofc)

[empty goblet]

-hugePages -maxmem=32768 -malloc=mimalloc_v214_lock_pages -maxFileCacheSize=16384 + lockpages enabled for the system and user account and the a3 process

[empty goblet]

could be someone eject and the plane gets hit or eject after hit ? ... {not sure what Redux does to planes}

[fast hornet]

possible.. my test was on our dev box, so i was the only client.. the plane just crashed after ejecting

[empty goblet]

arma, breaking the most advanced allocators since 2008, including those who were 'rock stable part of certain linux distro' 🤣

[whole cloud]

it was a crash on malloc

[heavy galleon]

I am just about to pick up new HW for Linux server, don't tell me I better get windows license for it too

[empty goblet]

307 is like 10 GB peak private byes and 10 GB peak working set, on Redux, peak virtual is 22GB

tho KOTH server with 413 peak virtual already is 38GB

[whole cloud]

Okey so the error I found, is in AI simulation.
But even if I disable all my AI tweaks.. it still happens.

Likely hitting an old bug here

[empty goblet]

that was joke related to some bug Suma found while testing dozens of allocators decade+ ago, for servers, Linux is superior

[empty goblet]

[whole cloud]

Ah the FSM deletes itself, while its running.. Yeah that'd be an old bug.... ugh

There is probably some new bug in latest prof there, but I cannot even get to it because it dies before then

[heavy galleon]

Good, the Ubuntu usb I made today will be handy then. 9700X time, hopefully will manage to not crash with ~130 players

[fair ore]

Personally, I think FSM is pronounced fism, like schism

[whole cloud]

The good old

_simulate = {_stack resize 0};
{_x call _simulate} forEach _stack

😠

int n = _stack.Size() - 1;
if (n < 0)
{
return;
}
This is the worst
if (_stack.IsEmpty()) return; that I've ever seen.

[empty goblet]

argh, 2nd server crashed ... w/o fulldump or mdmp ... just bidmp

Exception code: C0000005 ACCESS_VIOLATION at D645C4E6``` 🤣

it ain't enough to crash, let's double crash

[vivid reef]

same as my arma did just now. crash, and crashed again

[empty goblet]

80 players on KOTH, ~300 fps ... sitting in middle of combat zone, playable ...

[vivid reef]

how tf you make 300fps in middle of ao?

you got nasa super computers at bohemia?

[knotty wraith]

-- limitFPS

[empty goblet]

that's server fps ... my window has like 30 to 40 fps

[vivid reef]

good

i was happy about the +10avg fps and peak at 130 in ao with new profiling but the crashes aint worth it

[empty goblet]

ye the game is definitely smoother on the client ... so far my client didn't crash so that's +

[vivid reef]

lucky you

[plain trout]

KOTH+Zombies server?

[vivid reef]

it crashed after 25 mins, then restarted and crashed again immediately

[empty goblet]

no zombies anymore

but on the older machine i have different allocator ... the cma one

[spiral pond]

still getting the rendering issues with proxies (headgear, backpack, etc) on soldiers or as vehicle crew

[empty goblet]

ye i got some weird bug with stuff vanishing depending on angle of view

[spiral pond]

it seems like some frames it resets the relative position to 0

it appears to happen more if you turn your view on something, but also happens if you keep the view on something

video of that useful or no need?

[plain trout]

80-150fps at 2* 4k

[vivid reef]

maybe i was limited by the setcpu 12 core option from the last profiling i used

[empty goblet]

would be useful to see if it's same or different from what i got

[analog acorn]

I've recently had a couple of incidents of specific people being invisible depending on what angle you look at them from.....on stable though 🤔

[empty goblet]

so that makes 3 of us 🤣

[knotty wraith]

strangely enough, but my server is still working, there were personal crashes and often... but the server is working

[spiral pond]

OBS is glitching a lot tho - missing frames. probably need to do the pc restart first

still you can see it somewhat

[whole cloud]

Please don't rename the Arma exe, that makes crashes hard to analyze..
One crash is a similar malloc thing as the others. The others I don't know.

But I now don't get any more issues in YAAB. And the bug I fixed wouldn't cause this

[minor geode]

with the multithreading enhancements coming up and in place, is it necessary to change our CPU affinity still?

[whole cloud]

I don't remember it ever being necessary. Except on servers where you run multiple on the same machine and want to separate them

I think all the crashes i got so far, were profiling builds. Maybe only that is bad

[minor geode]

It had been an optimization suggestion for single core speed etc for so long, I dont even know if it's been valid tbh lol

[whole cloud]

Can't think of what that would help

[naive osprey]

There a way to properly launch Arma via the launcher when using the profiling build then?

Using the Steam profiling build hasn't allowed me to use commands like diag_captureSlowFrame since 2.14, so I've been manually downloading and renaming it.

[whole cloud]

no, you could use a symlink instead

probably all solutions are badin some way

[naive osprey]

Hm

Be a possibility to add an option in the launcher to switch which .exe to point towards?

[honest fulcrum]

Rename the original exe to "stable arma3_x64.exe" and then rename the branch exe to just good ol "arma3_x64.exe" is what I do

[naive osprey]

This is what he's asking me not to do when I provide crash dumps

[honest fulcrum]

Oh, sorry. I misunderstood

[knotty wraith]

I can't wait for the crash, it's time to sleep - the server is still alive

[whole cloud]

no

[minor geode]

But if you have the profiling build selected in Steam, it runs the arma3profiling_x64.exe by default right?

[plain trout]

steam replaces the arma_x64.exe if you run profiling

[whole cloud]

no, default is performance

[minor geode]

oh? so in order to actually run the profiling build, you should manually execute the arma3profiling_x64.exe?

[whole cloud]

If you need profiling, then yes

[minor geode]

after all these years, I never knew that. lol I just assumed Steam chose the right exe

[whole cloud]

It does choose the right exe

[naive osprey]

The "right" exe just isn't the one you expected

[minor geode]

right.. I know there's differences between the profiling and performance builds, namely some of the binaries, but with the profiling build be selectable now instead of how it was previously... it's good to get a clarification

[empty goblet]

you can always check .rpt file , it will tell you if you run profiling or stable (performance) build (at start of it) Type: Public Build: Stable Version: 2.18.152413

[whole cloud]

Okey found one AI thing that could cause issues, and shows up in one of Dwardens dumps.
🤞 that that's the crashes.

[vivid reef]

saw my dumps?

[whole cloud]

no not yet

Nice, the first crash on performance 😄

[vivid reef]

.HU

from mine dumps?

[whole cloud]

yeah. and its inside mimalloc too. Probably same thing

[vivid reef]

wait, performance is stable?

i switched to profiliing

[whole cloud]

Type: Public
Build: Stable
Version: 2.18.152413
That's what you're running

[vivid reef]

wtf

i had the profiling banner at the launcher

second crash stable too?

[whole cloud]

launcher always says that.
yes both same

[vivid reef]

wtf

[vivid reef]

Type: Public
Build: Stable
Version: 2.18.152302

why there is a different version number?

[whole cloud]

because thats not profiling branch, that is stable/default branch

[vivid reef]

that one is 19:05, the next rpt from the second one i send you

yeah, but the one you got should be profiling. the version number form the one you had in the rpt is different to the one i have now running stable

[whole cloud]

Yes the ones you sent are performance binary on profiling branch

[vivid reef]

now im totally braindead

Version: 2.18.152302 this was profiling in beta tab, the rpt i did send you

[whole cloud]

the two rpt's with crashes you sent me, are both 152413

[vivid reef]

yes

that should be profiling from beta tab in steam

[whole cloud]

it is yeah

[vivid reef]

good, but still im confused

profiling showing stable in rpt?

[whole cloud]

yes

[vivid reef]

oh

so you think its mimalloc?

[whole cloud]

the standard exe on profiling branch is a "Stable" build, just a beta stable one

no. It must be some wrong allocations done by game, but I couldn't reproduce any.
Only have one guess of one that could maybe happen

[vivid reef]

hmm

wasnt gaining that much more fps, but more smooth

[whole cloud]

Kinda weird, this AI thing probably causing the crashes, has been in since v6 and these crashes probably caused by it only appear today

[vivid reef]

[void badger]

v5 on Linux supremacy 💪

[empty goblet]

KOTH server is still running ... 80 to 90 players ... defying the crashes 🤣

[whole cloud]

152416 for tomorrow. With this the crashes should then be fixed.
I won't put up other binaries, you can run dedicated server with this too.
If there are no reported crashes with this we'll just push it tomorrow morning

[gusty viper]

Need client test with that binary too?

[whole cloud]

whoever crashed before probably

[vivid rune]

Is ingame the build number correct?

nevermind. I renamed the false exe

[empty goblet]

can't run on dedicated server, complains about lacking steam or something lol

[whole cloud]

oh right.. oh well

[gusty viper]

Crash while playing SP mission Combined Arms. Got dumps, will send

[vivid rune]

YAAB is running with two heavy mod sets and without.

I tried to play "lost signal" with a heavy modset and there was no animation flaws but without mods it still freezes.

[naive osprey]

Using this build (152416 symlinked as arma3_x64.exe), crashed ~5 minutes into playing Koth. Using MiMalloc. May be related to the Bad Conversion: string error happening nearly every frame, but still seems to be erroring inside the mem allocator.
¯\_(ツ)_/¯
https://www.mediafire.com/file/0gsbaef7hzw1d8l/arma3profiling_x64_2024-11-14_15-09-52.rar/file

[light cargo]

the launcher should really have support for specifying which exe to launch

[light cargo]

hmmmm, which alloc would be recommended for win7?

it seems i got improved performance by turning off ht and hugepages???

[heavy galleon]

perf still supports win 7?

[fair ore]

windows 8+

(windows 10 or 11)

[light cargo]

very funny
so, whatever default allocator then. system sucks and i cant run minalloc

[fair ore]

i mean it; 7 is four years past extended support.

It's hella eol

[hybrid cradle]

Bit like my knees

[light cargo]

im aware, yet im still using it because it has everything and will break if i upgrade to a newer os. i use legacy software here

[fair ore]

Then I guess you shouldn't expect support from anyone

[light cargo]

....

[spiral pond]

some chat commands like #monitor have auto completion. others like #captureframe dont - would be convinient to have the auto completion list updated 🙏

[woven loom]

Yeah, hardly anyone would still be using it at this point, so will be difficult for most people to know what stuff still works with it

[uneven kernel]

Fix for downloading at game start works 😄

Thank you dedmen

[tawdry gazelle]

I'm pretty fine Aha, though I didn't quite catch up the conversation. 😅

[plain trout]

Are you going to update mimalloc to v217?

[tawdry gazelle]

mimalloc handles thread safety issue internally, or it can't be a general allocator. For example, in C++, failed to handle thread-safety issue will cause new and delete malfunction due to free storage (or we call heap). 😂

[tawdry gazelle]

I assume v217 contains some minor updates and is not that complete, so I prefer waiting for a later version. 🤔

[whole cloud]

I didn't know there was auto completion in chat 😮

[plain trout]

been there all the time since OFP… 🙃

[woven loom]

So... YAAB ran without any problems for me today (but my game hadn't crashed so far anyway). No real change in performance when running the ultra preset + yaab standard settings, but i did see an improvement in ultra preset + 12km view distance. but i don't have a good record of this so not exactly sure by how much. I used to see around 70 fps with the latter settings iirc, now i saw 81, which is pretty nice.

But more interestingly, in of the scenes in YAAB there is a car that flips after an explosion. Previously, even last week, it would do a pretty short flip and very quickly fall back to the ground, almost in place

But now, it gets blown out of the scene and the physics looks a lot more natural

I haven't played a normal MP game in a long time though 😄

[empty goblet]

you can always use the cma allocator (the next best after mimalloc)

[light cargo]

thanks, is it stable?

[light cargo]

i understand, yes. but it's not a reason to whine against it.
the computer is pending for a replacement and it's grandfathered and the software will break with an upgrade which is unwanted

[light cargo]

amazing

[dusky perch]

yeah but no reason to call the other person a fascist

[light cargo]

its a expression for people who are too strict about things, so not really fascistness

[dusky perch]

calling someone n*zi is def a slur and is even punishable in alot of countries

[light cargo]

¯_(ツ)_/¯

[woven loom]

i'm super tempted to ask which countries these are lol

but that would be super off topic, so unfortunately we will never be able to question this outlandish claim

[tawdry gazelle]

have you tried mimalloc on Win7? will it work?

[light cargo]

uhhhhhh, but is it said it doesnt support win7 in the gh page?

doesn't it use win8+ api functions?

[tawdry gazelle]

Nope.

[tawdry gazelle]

Nope. Where did you read that? 😅

[light cargo]

(also technically win7 is valid to support albeit with more limited success because the game isnt going to drop win7 support soon)

oh oops

i vaguely recall people saying it doesnt work properly or something, sorry.

idk i didnt try it because of the potential of being unstable

[light cargo]

not easily googleable, im not sure what terms to put
i found something for linux not windows

[tawdry gazelle]

That could be true... I dont own an Win7 PC so I dont know if it works on Win7... 🤣

[light cargo]

my computer is a i7-4790

[tawdry gazelle]

So I ask if you tried it.😅

[light cargo]

thats ok, can always test a vm

yeah im going to try it

however im not a best case because my memory/motherboard/cpu is not at its best shape

[tawdry gazelle]

If the game started correctly and RPT shows allocator as mimalloc, then it runs well.

[light cargo]

ok, how about potential issues during gameplay?

it could bsod

[empty goblet]

i posted the latest download of CMA 2016 here #end_of_life_arma message

[tawdry gazelle]

Then we will see the crash log or BSOD info to see what caused it. Generally, CMA shouldn't issue a BSOD.

[light cargo]

ok

ohhh im not sure if intel compiler doesnt support win7 hmmmm

we'll see, i'll give it a try!

ill try general use first

i have 16gb of ram but as i said huge pages does slow me down

i think i just recall that mimalloc kinda doesnt support win7, just that

im not sure however

also i have significant issues with my installation such as having a large non-paged pool occupied by the registry

Well, it causes registry operations to take a longer time than usual to complete

also @tawdry gazelle, mimalloc might be due for a update since its 245 commits behind microsoft?

okay im launching, so far so good

[whole cloud]

Found one more rare use-after-free, I could reproduce it in the mission where @gusty viper crashed.
And that one, is actually new code that came with v9, maybe its that

But trust a "maybe hopefully that was the cause" over the weekend? ehh

[plain trout]

people can always switch back to vanilla?

[whole cloud]

Yeah after they'll notice its unstable by repeatedly crashing mid mission

[light cargo]

oof, i cant upload images here

[spice dirge]

upload them into some stupid discord bot DMs (like MEE6 or whatever), then post the link here 😄

[light cargo]

discord cdn expires after 24h

[ashen helm]

you can forward messages to bypass CDN expiration.

[spice dirge]

#arma3_config message going strong 2 years after

[ashen helm]

Like this

[silk pewter]

or like that

[light cargo]

uh, it wont allow forwarding attachments/links if they're restricted in the target channel

[spice dirge]

literally a link to the image posted in literally MEE6 DMs

[naive osprey]

Imgur link embed too ¯_(ツ)_/¯
https://i.imgur.com/Ye63DcL.png

[light cargo]

lol ok

finally

ok, this is the second run of yaab, same fps as last time, and this is also modded lol.

also this is with mimalloc

[inland dew]

iGPU?

[light cargo]

one of the mods is blastcore

gtx 1660

yea, the system is actually pretty hampered

i am using a better smoke mod and blastcore murr

so that's probs the lag and the view distance

[light cargo]

kinda abusable, i'd expect a patch soon when people figure that out

anyways i have cup, real engine, niarms, ace with tracers, particles and no medical as well

third run gave me 16.4 average fps

max 31 fps, min 5 fps

[silk pewter]

I don't get it, are we testing mod compatibility now or just vanilla?

[light cargo]

i was testing mimalloc with my usual mods :P

im going to test with 2 cpu count, also this is with HT off

[autumn timber]

Hmm... I had been using a i7-4790k with a GTX1070 for years and I could swear that I had better FPS.
||And yes, obviously, Windows 10, which has always worked fine and still works fine for my father who got that computer from me when I switched to something better||

[light cargo]

@tawdry gazelle excuse me, are there benefits with ht in mimalloc and lower cpu count?

[light cargo]

its a 4790k and hdd, also with some sound mods i think.
ehhh its whatever, the system is paging constantly which is why its slow

splendid smoke rework is enabled

so yeah... :D

i'll try yaab with vanilla

[light cargo]

what are your video settings btw?

[plain trout]

#hardware_vs_arma

[light cargo]

sorry

its related to here because of the use of performance branch and mimalloc

with -cpucount=2 it freezes after loading a mission such as yaab

it wasnt freezing for so long without it

[autumn timber]

You must have missed the part where I said that I gave that computer away to my father (click the spoiler 😛 ), which was 2 years ago, so my current settings are irrelevant and i don't remember the old ones

[light cargo]

sorry lol its ok

i adjusted some of my settings and in vanilla im getting ~26 fps

better fps now, yay.... well i'll leave it there

[inland dew]

hdd...

[autumn timber]

As long as the game fits into memory, that shouldn't be a problem I think? (don't quote me on that, though)

[inland dew]

that's where so so low fps comes from and the freezes

nothing to do with mallocs or perf branch

[hybrid cradle]

no such thing as good arma performance on a hard drive

[inland dew]

thus #hardware_vs_arma

[light cargo]

yeah uh, it does improve on a second run however. yeah sound and graphics mods do impact my performance a lot more

[inland dew]

#hardware_vs_arma

[light cargo]

whatever, at least im more comfortable with my game now

[hybrid cradle]

step 1: upgrade to an ssd
step 2: get windows 10/11

[quaint flame]

At this point. Buy a new pc. 4th generation is super slow

[empty goblet]

step 5 , realize Windows was mistake and migrate to Linux OS

[spice dirge]

mandatory path capitalization joke

[light cargo]

yeah, im aware, soon™

[knotty wraith]

I used 4790 k + m2 at one time

[light cargo]

nice, i do plan to have a 2tb m2 which is pcie 2.0 as boot drive for this system soon. but financial issues

[hybrid cradle]

Mandatory path capitalization joke*

[whole cloud]

⚠️ Hotfix pushed to 152418
another attempt at fixing the crashes

[silk pewter]

mandatory Path capitalisation joke** 😁

[hybrid cradle]

😦

remember to remove the french language pack

[silk pewter]

en-GB is still a thing

[silk pewter]

(wait that was a rm -fr / reference, right 😄)

[hybrid cradle]

yes 🙂

[silk pewter]

thought you were nudging at the capitalisation, sorry for not catching it ^^

[hybrid cradle]

i wish i could remove my ability to speak french 😦

[silk pewter]

https://tenor.com/view/non-french-no-french-monkey-no-to-france-french-flag-gif-21517493

anyway 🤣

[hybrid cradle]

😦

[heavy galleon]

Might have been mentioned few times.
I was stress testing our new server HW and was getting flood messages about me in the server console, then my game crashed.
I spawned around 90 grads and made them all shoot 20 rockets at the same time.
Any recommendations to avoid the flood warnings etc?

Both client and server are on perf, server is Linux.

[empty goblet]

ok, let's try again , with 418

[fast hornet]

i just hope that ancient Win7 system you mentioned is not connected to the internet. Otherwise gtfo

[light cargo]

ᕙ༼ຈل͜ຈ༽ᕗ ͡° ͜ʖ ͡°

[heavy galleon]

So for system with 8cores 16 threads you recommended setting -cpucount=8 and to not set -enableHT, right?

[hybrid cradle]

let engine autodetect

[inland dew]

just do -enableHT and that's it

[whole cloud]

if you set cpuCount, then enableHT does nothing.
The correct answer is, set nothing

[heavy galleon]

So neither cpuCount and enableHT, or only enableHT?

[silk pewter]

n o t h i n g

don't use anything 😄

https://tenor.com/view/nothing-to-see-here-gif-13539524016057276385

[heavy galleon]

so ./arma3server_x64 -cfg="$networkConfig" -config="$serverConfig" -name="$profileName" -mod="$mods" -servermod="$servermods" -limitFPS=75 -nosplash -loadMissionToMemory -bandwidthAlg=2 -maxFileCacheSize=8192 -hugepages world=empty -port=2302 -noSound 2>&1 | tee logs/arma3Server_$(date +%F_%T).rpt
sounds good then

[inland dew]

#perf_prof_branch message

[heavy galleon]

but they said not a thing

[inland dew]

ht is fine, if you don't use cpucount

[whole cloud]

-bandwidthAlg=2
2 is default on profiling branch

[hallow lantern]

no more crashes on linux but

waypoints are no longer working, units are stuck and won't move ( but they will fire eventually )

[whole cloud]

Might be the animations bug thing

Sometimes on linux, animations get bugged and everyone is stuck in dragging animation and cannot move around (but can fire)

[hallow lantern]

hmm that never happened as far I can remember, but if I just place an unit with zeus and give it the move waypoint, it just stand there

( I'm testing with lambs, without lambs there are no crash )

https://streamable.com/mkwtse

aaand HC just crashed ( but not the server lol )

[indigo beacon]

Oddly enough on the new update, borderless windowed mode tanks to like 6-9 fps, going to fullscreen puts me back to normal expected FPS

[hallow lantern]

tried replay several tested and working missions, with or without HC, noone follow waypoints of any type

[gusty viper]

Time to die in mission Combined Arms again!

[empty goblet]

you would be suprised but MS still releasing patches for it, just not for normal users ofc ... and from perspective of holes, there was so many RCE for 7,8,10 and 11 that it's just like tossing coin to get hit ... better to just move on nix 😁

[empty goblet]

i still use it, it's hard to break my habbit 🤣

anyway , so far no crash on servers

[fast hornet]

think Win7 is even out of the extended paid support by now.. it's that old

[gusty viper]

Died succesfully, no crash this time around @whole cloud

[empty goblet]

ye it's old but it was simple and working system, even 8.1 was damn good (if you ignore the early UI woes) the kernel was fine

[gusty viper]

(That mission is a crazy ass infantry action)

[silk pewter]

*clenches to his Windows 98 FE*

[gusty viper]

Also: in these SP mission everyone is following WPs as normal.

[fast hornet]

sure.. i loved my Windows 2000 as well.. ran perfectly. But at some point you'll have to let it die in peace 😂

and NT 3.5 before that..

[empty goblet]

sorta but W7 just running games (in sanboxed network or w/o network) will keep running games until it the hardware dies physical way as long the games aren't updated in way to not run on W7 🤣

and like i said if you want something what keep on going just use linux ... 🤣 because i honestly don't even want to try think how W12 pans out after 11

[gusty viper]

Died succesfully in mission Infantry too, no crash. Seems fixed for me. Was there a recent change that npc's throw more grenades and use underbarrels? Sjees!

[vivid reef]

3.11 running on dos 6.2 good ol time

[fast hornet]

ofc, without a network connection everyone can do whatever they want.. plenty of retro gaming machines exist.
But it sounded like the guy is running Win7 as his main system with internet connection and browser and everything.. and that's just an insanely big ahole move on his side. That thing does not belong on networked systems anymore

[empty goblet]

well if he behind good firewall and NAT he is fine ...

if you run any Windows including 11 on DMZ on internet, you will likely be pwned sooner than later

[light cargo]

um not true since jan 2023 for win7 and jan 2024 for server 2008r2, but there's still updates for embedded standard

[empty goblet]

like i said, updates for the W7 still exist ... even for W98 if you know where to look (thanks to certain terminals using them)

[light cargo]

just not over wu/wsus?

[empty goblet]

anyway ... 90 player son Redux WL, 120+ fps ... 50 players on KOTH and some on the other Warlords servers, so far no crash ...

i do wonder if the Z fighting got worse or it's just me for long range edge shimmer

[kindred radish]

Z fighting at longe range was always bad, as long i can remember. Main issue imho is the models, some parts of the building are to close to others.

[autumn timber]

I'm too lazy to search for it, but I bet there are plenty of ways to hack win7 even when "behind a good firewall and NAT".

Like probably some JS code that would connect to localhost from your browser or something like that, maybe an image with the right localhost url, maybe do something fun with the local printer, etc...

[empty goblet]

js code is problem of browser, so use browser which is still supported like Firefox ESR or one of the firefox forks

and to get pwned via js doesn't stop on platform , there are literally weekly 0days for chromium for past 12 months

[spiral pond]

#360648487378878464_XXX

[empty goblet]

anyway , we moving away from the #perf_prof_branch purpose 😁

[autumn timber]

But these 0days get patched, unlike the 1000-days for Win7. That's the point

[empty goblet]

no serious hacker cares about w7 at this point ...

[fast hornet]

security by obscurity never worked and never will.

[autumn timber]

I don't want to drag the offtopic discussion further, but... I'm sure they do. Just as there are ALWAYS attempts to scan the ssh servers for admin:admin, etc...
That's just low hanging fruit reaper that you can leave running on

[knotty wraith]

where is the administrator? 😆

[empty goblet]

hmm, i guess i got another crash in ui by dragging / clicking stuff 🤣

WL redux server, average ~70 fps for full 100 players 17:59:31 Server load: FPS 69, memory used: 5012 MB, out: 36769 Kbps, in: 1253 Kbps, NG:0, G:6441, BE-NG:0, BE-G:0, RQ:0, Players: 100 (L:1, R:1, B:2, G:95, D:1), JIP (T:0 Q:73458) and no crash yet

[silk pewter]

https://tenor.com/view/nod-kermit-nod-nodding-gif-24624072

[light cargo]

and this is why changing default ports does actually work.... most dont care to scan systems and just try on default ports

¯_(ツ)_/¯

moving on

[light cargo]

wrong, just try exposing a service on the wrong port and see how bad the attempts are for the incorrect service 🤣

enough said now!!

um, what priorities are there for perf/prof currently?

[heavy vortex]

Making all the recent changes work without bugs, I imagine.

[gritty wasp]

Use Linux they said

[heavy vortex]

That animation bug was very commonly replicated by running your Linux server with the advanced rappelling mod.

[heavy galleon]

Better than paying 35$ a month for license

[drifting swan]

👍

22:00:31 Server load: FPS 90, memory used: 3544 MB, out: 42867 Kbps, in: 1304 Kbps, NG:0, G:0, BE-NG:0, BE-G:0, RQ:0, Players: 200 (L:0, R:0, B:0, G:200, D:0), JIP (T:0 Q:52766)

but second server:

21:04:14 Server load: FPS 91, memory used: 2936 MB, out: 10970 Kbps, in: 406 Kbps, NG:0, G:0, BE-NG:0, BE-G:0, RQ:0, Players: 198 (L:1, R:1, B:0, G:196, D:0), JIP (T:0 Q:24975)
21:04:19 Server load: FPS 91, memory used: 2937 MB, out: 4313 Kbps, in: 324 Kbps, NG:0, G:0, BE-NG:9, BE-G:0, RQ:0, Players: 198 (L:1, R:1, B:0, G:196, D:0), JIP (T:0 Q:24982)
21:04:24 Server load: FPS 83, memory used: 2953 MB, out: 18615 Kbps, in: 1283 Kbps, NG:0, G:179635, BE-NG:0, BE-G:0, RQ:0, Players: 198 (L:0, R:2, B:0, G:196, D:0), JIP (T:0 Q:25064)
21:04:29 Server load: FPS 85, memory used: 2968 MB, out: 38783 Kbps, in: 2796 Kbps, NG:0, G:50895, BE-NG:0, BE-G:0, RQ:0, Players: 198 (L:0, R:2, B:0, G:196, D:0), JIP (T:0 Q:25789)
21:04:34 Server load: FPS 44, memory used: 2986 MB, out: 45466 Kbps, in: 1068 Kbps, NG:0, G:2862, BE-NG:0, BE-G:0, RQ:0, Players: 197 (L:0, R:1, B:1, G:195, D:0), JIP (T:0 Q:26326)
21:04:56 Server load: FPS 1, memory used: 2987 MB, out: 1310 Kbps, in: 22 Kbps, NG:0, G:1269480, BE-NG:108, BE-G:0, RQ:29740, Players: 197 (L:1, R:0, B:1, G:195, D:0), JIP (T:0 Q:26346)
21:05:11 Server load: FPS 0, memory used: 2991 MB, out: 3903 Kbps, in: 66 Kbps, NG:0, G:2771929, BE-NG:378, BE-G:0, RQ:55467, Players: 197 (L:2, R:1, B:1, G:193, D:0), JIP (T:0 Q:26356)

I had to roll back the second server to v5 to play :)

[empty goblet]

did you do #captureFrame when that fps went to <1 ?

[drifting swan]

nope

[heavy vortex]

That second server isn't 32-bit, is it?

[inland dew]

would have been strange if it was

[empty goblet]

you should to know what went wrong

[drifting swan]

no, all servers is x64

[drifting swan]

to be honest, i thought captureFrame is only available in diagnostics exe :D

[empty goblet]

and profiling

anyway no server crash so far, so i guess this build is doing well ... and now ... i'm going to take some nap

[drifting swan]

Oh, I started the servers via arma3server_x64, but should be via arma3serverprofiling_x64?

[fast hornet]

depends on what you want... the normal one is perf and the profiling is, well, profiling

[inland dew]

profiling is for #captureFrame

[drifting swan]

ok, got it. we were just playing and didn't think we'd need to capture frames

[tawdry gazelle]

mimalloc has nothing to do with ht or cpu count. they are irrelevant.

[uncut linden]

Oh yeah, I remember writing a 'fix' for that. Is that still a thing?

[coral pendant]

So this new things gives better frames?

[eternal kraken]

found a visual glitch bug yesterday in our playsession with v.2.18.152.418 Stable

every client and server running v.2.18.152.418 Stable

i recognized that bug already before in earlier builds ( first time in v08 (374) i think) but could not capture it on video until now

[empty goblet]

no server crash yet ....

[empty goblet]

ye i observed some 'strange' lil visual glitches

[light cargo]

well, its just weird that its freezing when using mimalloc and cpucount 2
also setting those in a specific way is supposed to yield more performance apparently

[light cargo]

uh, this channel is for the performance/profiling branch of the game,

[dusky spire]

No more flickering with the new versions, thank you & well done, amazing work 🤘

[gusty viper]

No, the advice of the devs here is to not set those things, unless you try to run multiple server processes on the same machine. Are you?

[light cargo]

.....
i do that sometimes but ds+client

[gusty viper]

Yeah that makes sense, can help fps on the client quite a bit with a dedicated server process doing the heavy work. But recent improvements also make it possible to run just with client.

[kindred radish]

removeUniform _unit executed on the server, causes this spam (Unit was spawned and equipped by the server and then player controlled).
100+ lines in 7 minutes just for the one UniformContainer.

// 2:523 = netID of UniformContainer
2:33:41 Server: Object 2:523 not found (message Type_466)
12:33:46 Server: Object 2:523 not found (message Type_98)
12:33:47 Server: Object 2:523 not found (message Type_466)
12:33:51 Server: Object 2:523 not found (message Type_126)
12:33:52 Server: Object 2:523 not found (message Type_466)
12:33:57 Server: Object 2:523 not found (message Type_466)
12:34:02 Server: Object 2:523 not found (message Type_466)
12:34:06 Server: Object 2:523 not found (message Type_98)
12:34:07 Server: Object 2:523 not found (message Type_466)
12:34:12 Server: Object 2:523 not found (message Type_466)
12:34:17 Server: Object 2:523 not found (message Type_466)

[whole cloud]

Yeah right. So why are you here then?
I'd prefer to stay with your stuff, and not tell people posting valid things in the correct place, that they are in the wrong..

[empty goblet]

19:01:20  ➥ Context:     [] L5 (mpmissions\__CUR_MP.Altis\client\*c_fl_mapDraw.sqf)
    [] L5 (mpmissions\__CUR_MP.Altis\client\*c_fl_mapDraw.sqf)
    [] L5 (mpmissions\__CUR_MP.Altis\client\*c_fl_mapDraw.sqf)
``` hmm, noticed this spam in  2.18.152418 client RPT.  just 712 657 times ![Armad](https://cdn.discordapp.com/emojis/782218958954037280.webp?size=128 "Armad")

[light cargo]

wdym? im so sorry for this

[empty goblet]

hmm, just joined game and got this 13:51:22 O Bravo 1-6:1 (Dwarden) (Team Leader): Unknown action ListRightVehicleDisplay, target O Bravo 1-6:1 (Dwarden)

[ashen helm]

Unit flickering is gone but the LOD loading is very rough compared to stable. First the head flickers, then the vests and backpacks. Definitely not tied to my hardware as it used to not do that unless I was scrolling through arsenal items very fast.

[empty goblet]

also i had weird case of very slow loading of KOTH on the client since 418 but happened once so not sure if it was fluke or indicator of something weird after roles selection/entering mission

[spiral pond]

something seems wrong with #captureslowframe

i can get it only to trigger with #captureslowframe total 0.001 (above 1 ms) not with #captureslowframe total 0.01 (above 10 ms)

the problem seems to be there is two total for the server

so it doesnt check the real total main thread - and the other is just very short

---

https://community.bistudio.com/wiki/diag_logSlowFrame seems to work now in A3 - intended?

[spiral pond]

probably a wrongly labels scope or it uses the total label due to some coding mistake

[empty goblet]

something wrong with corpse removal #warlords_discussion message if true i don't get how come corpses aren't removed after 30 minutes (if new corpses are produced all the time)

[ashen helm]

also I believe this happens only once per a newly "encountered" unit. Ones I go through all the LOD checks, then the LOD change happens without said glitch no matter how fast I apprach the unit (or from whatever distance))

[empty goblet]

i do wonder about the corpse/wrecks/weaponholders garbage collector ... do we have anything to track when was last interaction with the corpse / weaponholder ?
because if we can, we shall add this to the manager .... to remove stuff which wasn't interacted with, after defined time ... (ideal to resolve case of GC fails ie spawns where players idle (triggers minPlayerDistance[]))
if we can't track the last iteration then something like max lifetime irrelevant on minPlayerDistance and separated lifetimes for respawned vs non respawned
feel free to tell me this is isn't needed but would solve some problems 😁

[gusty viper]

I also saw this in various SP missions I played yesterday with the latest perf binary. It’s nuanced but in particular heads can appear to flicker sometimes.

[feral harness]

We dont have this value, and to add it it has to be global so any client can change it. Thats a bit of work. We already have priority value calculated which item is deleted first, this could add to priority, however I'm not sure what are we trying to solve here?

Please state a concrete example of a problem

[empty goblet]

atm. if you have player which doesn't move, the item stays ... so what's needed it way to break out of that situation

[feral harness]

Then it stays until max time and it will get removed

Ah, it does not. Can make it so that if (time > maxTime * 1.5) remove anyway even if close to player. How abouot this for solution?

so if player moves away within additional 1/2 of maxtime it gets deleted, if not, it gets deleted after additional 1/2 maxtime is passed

or make it maxtime + mintime as absolute maxtime

i like that more

[empty goblet]

also it needs to be definable for each type (it to be able define different timers 1 = All vs 2 = None_But_Respawned vs 3 = All_But_Respawned

weaponholderRemovalTotalTime[]={respawned_timer,not_respawned_timer};
wreckRemovalTotalTime[]={respawned_timer,not_respawned_timer};

[feral harness]

However we are talking about about a few items that are not cleared because player is near and doesnt move, how bad could it be?

[empty goblet]

in fact i really wish all the GC values were possible to define for respawned vs not respawned with separate values, because now it's just either 1 all or one of them, but no solution for both

[feral harness]

Im just gonna say no to any additional params in. Im not doing it it is already way too complicated

[empty goblet]

we talk about gameplay cripling problem as just tested on WL Redux server with 100 players after 1-2 hours ...

while server happily crunch at 100 fps, there is so many non-removed items around the 2 sides bases(spawns) that the performance is unplayable for the clients

[feral harness]

you described a concrete problem, the player doesnt move it doesnt get deleted at all, I am giving you a solution that will work and is not going to be difficult to implement. I am not doing new params. Ask dedmen if he wants to do it

[empty goblet]

if we could define timers for both spawned and not spawned items (AI) with separate value and then also define the total timeout for each ... that would solve it

[feral harness]

there is always scripting solution in addition.

[empty goblet]

it's not just one player afking, the players keep respawning (death,jip etc)

[feral harness]

so you want my solution or not?

[empty goblet]

that's why i asked and suggested the timers separate from spawned and not spawned items Manager Mode 2 and 3 at minimum

[empty goblet]

yes if the values can be customized ^ , do i make sense why ?

[feral harness]

you can pseudo customise duration by defining min and max time as it is going to be the sum

[empty goblet]

it would be better if this is separately defined values otherwise it will be again limited by trying to figure some optimum for 3 different cases

[feral harness]

you define manager mode at the same time you define min maxtime, what are you saying? Why do you need different times params?

[empty goblet]

ie i'm forced to use the 1 = All , that means AI items vs respawned player items already using same value (because there is no way to define values different for each of those 2)

[feral harness]

Each item type has own manager mode

and the times you define will work for that mode

[empty goblet]

yes but the manager mode is 1, 2 or 3, i can't define timeout A for 2 and timeout B for 3

[feral harness]

changed mode? change times

[spiral pond]

from what i understand Dwarden asks for different duration of AI vs player bodies, plus their weaponholders

for wrecks/vehicles i dont see how you can make that distinction tho

[empty goblet]

and wrecks from player vehicles vs wrecks from AI spawn (if that can be determined)

but ofc prio on the weaponholders and corpses

[feral harness]

weaponholderManagerMode = 1; weaponholderRemovalMinTime = whatever <- min time for mode 1
weaponholderManagerMode =0; weaponholderRemovalMinTime = whateverever <- min time for mode 0

the times are for the mode you chose

[empty goblet]

but i need it for BOTH each different

[feral harness]

changed mode, define different times for it

[spiral pond]

is there even a flag to determine if a body/WH is from AI or a player. probably not

[feral harness]

i dont understand, you cannot have weaponholderManagerMode = 1; and weaponholderManagerMode = 0; at the same time

[feral harness]

That will require a new remains collector entirely

[spiral pond]

what about GC based events/EHs - would that be more feasible?

aka make a scripted system/adjustments based on the engine GC system

ie via Killed EH you can store if AI or player origin on corpse/WH/vehicle easily
if you then had an EH on maxRemovalTime reached, you could do a forced scripted removal if was AI/player (based on the desired system)

[feral harness]

I can add mission EHs to remains collector

[empty goblet]

in relation to above, after discussion with KK, we try minPlayerDistance[] = {corpses, wrecks, weaponholders, corpsesforceremovalmaxtimemultiplier, wrecksforceremovalmaxtimemultiplier, weaponholdersforceremovalmaxtimemultiplier} default, 0 is infinity

while talking to KK i missed there is 140 players on KOTH server

[empty goblet]

it's bit laggy but the server FPS isn't the cause 18:18:22 Server load: FPS 133, memory used: 4335 MB, out: 64705 Kbps, in: 1406 Kbps, NG:0, G:0, BE-NG:0, BE-G:0, RQ:0, Players: 135 (L:0, R:3, B:10, G:120, D:2), JIP (T:0 Q:18938)

[empty goblet]

this error seems to be some username causing it ... ie unicode character !? ... each time i open map / minimap / any map (like the death one or spawn board)

[feral harness]

could be copy paste from forums, it drags along some wierd non printable characters

[feral harness]

I would want to see the size of jip queue after dedmen adds my commit to profiling

[empty goblet]

that 135 has exported jip queue ,same as the 100 player redux ... plus captureFrame on each, this is server side ofc

[tribal pasture]

I wonder what are these:

Type_291 :  3372
Type_248 :  5623

[vivid reef]

Current profiling is running smooth af

[whole cloud]

That makes the second report of that.
Must be something caused by the mission 🤔

[whole cloud]

Oh yeah I noticed total not working on server, but didn't know why. I just use sLoop on server, that is why then..

[ashen helm]

another thing I've noticed, not all difficulty settings are being saved. Vision aid toggles itself on after being disabled either in main menu or during a scenario. Reduced damage seems to be toggled off but for example RHS prompts about it being enabled anyway, so I assume only the UI change happens without actually disabling the setting

Edit: creating a new profile seems to fix this, as in it does not happen on new profiles.

Dedmen please quit having fun and fix!!!

[ashen helm]

does anyone have the previous v9 before it was hotfixed? Wanted to see if this issue happens there too

[empty goblet]

at certain moment the server started to show 'jitter' desync on player movement and it didn't fix even as playercounter decreased

[naive osprey]

Pretty sure it was hotfixed twice, but I could be wrong - Here is 152413 - Current is 152418
https://www.mediafire.com/file/br32x7z83u7hrxb/arma3profiling_x64_v9_152413.exe/file

[ashen helm]

thanks

[empty goblet]

meow , this rendering bug ... driving at quad and it hide the tree infront of me 🤣 appeared at last second

[tawdry gazelle]

mimalloc has updated to v2.1.7: https://github.com/GoldJohnKing/mimalloc/releases/tag/Arma-3-v2.1.7-20241117

[jagged swan]

is it verified by BattleEye ?

[tawdry gazelle]

I guess no. It has just published less than 5 minutes.😅

[jagged swan]

so should we wait for half of the year to let them verify this version of malloc ?)

[tawdry gazelle]

I dont know, it depends on BE...😅

or you may submit a request to BE.
https://www.battleye.com/contact/

[hybrid cradle]

@whole cloud iirc you could whitelist/similar? ^

[fair geyser]

Seem to be getting worse performance (alot worse) when firing effect heavy weapons (Firewills A10-C) than I did before using the profile build

FPS tanks (1-2 FPS) until the effects are gone, while without the performance build it only (more or less) halves my FPS (30-35 FPS) until the effects are gone

Not using blastcore or anything like that

Particles are set to low

[spiral pond]

can you make some frame captures when you have low fps

howTo:

1. run game with arma3profiling_x64.exe (either rename it to arma3_x64.exe before you start with A3 launcher, or just bat file with arma3profiling_x64.exe and your startup parameters)
2. play as usual
3. when FPS is low
4. hold left shift plus - (minus key) on numblock
5. type frame
6. frame capture UI opens
7. hit "copy all" button in the bottom
8. open a text editor, paste the code (ctrl+p) and save file
9. drop in here

[uneven kernel]

This should be pinned somewhere

[lethal forum]

wow permonace is amazing im ona widescreen monitor dld Arma 3 again to test and im impressed, could you guys help me with fov settings for this res? fov calculators are off

[lethal forum]

5120x1440 sorry

[vivid reef]

any info when the reload-magazin-poof bug is fixed in stable client?

[gritty wasp]

3 weeks ago
https://dev.arma3.com/post/spotrep-00116

[vivid reef]

apparently it still happens

[patent dome]

Then someone on the server has not updated their game, probably.

[vivid reef]

by now everyone should have updatet client, im trying to figure out if our servers are hotfixed

[empty goblet]

can someone please check 2.18.152418 new Garbage collector ? seems it's not cleaning #warlords_discussion message (ignore the part about spawns, but sectors failing to clean w/o players nearby indicate something going wrong)

[empty goblet]

just use the requiredBuild in server.cfg to enforce it, otherwise it will be always chance someone breaking things

[vivid reef]

thats what roid doesnt want to

[empty goblet]

?

add this to server.cfg, to ensure no pre-hotfix client join and breaks the game for others
requiredBuild = 152302; // this is the hotfix build

so your server should be 302 or 307 (that's the last profiling before those experimental MT improvements)

[vivid reef]

im not in charge of the servers, thats roidrage from c4g

he doesnt want to force a build

[empty goblet]

well then he is acting stupid because that's recommended to do so until 2.20 released

anyone who joins with unpatched older binary (be it RC 2.18 or pre-hotfix 2.18 or older profiling before 302, will cause issues)

[empty goblet]

Redux server crashed, reverting to 152307

[light cargo]

sorry you're sol then, good luck convincing them to put it in for the given reasons

[vivid reef]

not going to convince anyone

[light cargo]

too bad, switch providers or wait til next version

[knotty wraith]

my observations of AI and HC... At v9, AI units (not players) begin to freeze in place (running in place, no damage) when more units spawn and at the moment a new player enters the game. Today we rolled back to v5 - there are no such problems at all. I don’t know what the problem is: in the hardware or in the width of the Internet channel, but at the moment there is only one conclusion for my server - v9 is good for clients, for a server with AI and HC - version v5.
v9 had a crash - Deadman dumps

[real creek]

I have a feeling some people will see their YAAB results go up a little from the workshop bug fix, I always used to only measure from 2nd run onwards but some people probably had smaller hitches that were difficult to perceive

[light cargo]

interesting, we have a similar issue and it turns out that the mission is very complex and the network settings do suck.
not sure if they tried downgrading for a complex mission that we attempted to play like 3 times

[random isle]

also 8 out of 10 times when you hit hc ai with a car car will come to dead stop or in many cases will get pushed away or even launched i had this going for years

[shell lion]

Yesterday switched from v5 to v9 with my server and HC (2.18.152418), before the Saturday night mission and after a short time when many units are running on the HC massive FPS drops down to zero (freezes).
Only on the HC.
Startupparams enable HT is set.
HC runs on same machine but different CPU counts as the server.
AI units only run on the HC.
No new RPT data entries that can help.
Error could be fixed by kicking the HC.
Now everything runs on the server without any problems.
Number of players approx. 20, number of units approx. 150-160.
PvE only. Jet, helicopter in the air.
Maybe a problem with client (HC) and server on the same machine with the massive fps drops?
All previous version before v5, no problems on server and HC.
I've been using v9 as a client since it came out and only have the unit flickering issue with players so far, otherwise no FPS drops.

[knotty wraith]

Where have you been all this time my friend)

[bold raven]

I think ur developing Arma 3 in ur brain right now. Straight cooker

[restive pilot]

Well I for one don't even understand what you're saying because you're using a bad translation service

[runic sigil]

Encountered this issue recently
https://feedback.bistudio.com/T125049
Appears to be incomplete dik codes for F13~F24
DIK codes only has up to f15, higher displays as "UNEXPECTED_KEY_ID"
Issue seems to be noone added the rest of the scancodes to the file

Probably could be fixed at least partially by adding to "\a3\ui_f\hpp\definedikcodes.inc"
Might need stringtable updates too, which might be more work than desired

#define DIK_F16             0x67
#define DIK_F17             0x68
#define DIK_F18             0x69
#define DIK_F19             0x6a
#define DIK_F20             0x6b
#define DIK_F21             0x6c
#define DIK_F22             0x6d
#define DIK_F23             0x6e
#define DIK_F24             0x76

@whole cloud
Pinging because it might be an easy fix that just hadn't been done
Here because this is the channel where current development on Arma 3 is mostly being discussed
If it isn't easy to fix, it's a minor annoyance issue that doesn't matter much

[naive osprey]

I love my UNEXPECTED_KEY_ID keybinds though! Its a 1 in 9 chance I hit the right key, perfect for gamblers!

[broken phoenix]

Question for the people that run the LAMBs mods on the perf branch, has anyone experienced LAMBs killing the servers randomly with
File /z/lambs/addons/main/functions/UnitAction/fnc_doFleeing.sqf..., line 73
If I remember correctly I heard there was an issue in the recent past, but unsure if this is related to that. perf v9

[chilly geyser]

As I remember, dedmen fix this issue approx 3 or 4 days ago

[chilly geyser]

Can u please share basic.cfg?

I didn't find any info about server network parameters and don't know how i can configure it correctly

[harsh grotto]

https://imgur.com/M5SZeZq I think it's not a very good FPS in the lobby

[spiral pond]

depends whats rendered in the background

is this from main menu, or from mission back to lobby?

[harsh grotto]

from main menu

ill try reinstalling Windows, hope it maybe will fix everything.

[spiral pond]

what are your A3 startup parameters?

[harsh grotto]

-skipIntro -noSplash -name=[STC]yod -exThreads=7 -malloc=mimalloc_v214_lock_pages -hugePages -world=empty -debug -mod= -beservice

[spiral pond]

-world=VR -skipIntro should give you max fps there

hm interesting

next time drop the log from that capture here

[harsh grotto]

i can drop in DM

[spiral pond]

sure

[harsh grotto]

sent

[spiral pond]

there is no private/hardware/game info that log tho. so unless permissions dont allow dropping it in here, there is no harm. if permissions dont allow, just use some pastebin site and drop the link to that

[harsh grotto]

I know, Im just too lazy to upload to pastebin )

if you want, you can send it here if you see fit, i just dont have the rights to do it

[spiral pond]

not a very good FPS in the lobby
lesson learnt is that viewdistance setting matters also in menus. aka dont set too high [ie max 5k] (especially in MP where you dont need to [usually])

[plain trout]

Yes. Performance drops from around 10-12km. I mentioned that somewhere a few days ago. It’s on perf and vanilla. The fps is worse in the empty world of the menu compared to EDEN/ingame.

[spiral pond]

it seems to be caused by water rendering

maybe the dummy world has water somewhere or reaches infinite terrain - which can cause issues

[lethal forum]

Been playing around on the KOTH test server fps is way better then usual but i notice the same problem as in the regular branch when I get into town where the action is my gpu is running at 20% im on a 4090 so instead of running at stable 2800mhz it goes all the way down to 900mhz-1800mhz which brings down the frames.

[quaint flame]

That's because your CPU is the bottleneck. The card reducing its frequency is just normal behaviour.

[void vine]

It's a visibility check issue

[whole cloud]

need repro

[whole cloud]

editing game FOV is not supported.

[whole cloud]

Sorry? This is #perf_prof_branch ? Were you looking for #arma3_feedback_tracker ?

Here because this is the channel where current development on Arma 3 is mostly being discussed
No. This is the channel where #perf_prof_branch is being discussed.

[whole cloud]

post the full error?

[broken phoenix]

 2:56:33 Error: Mismatched Begin/EndContexts
note: Minidump has been generated into the file C:\Users\Admin\Documents\Faster\ArmA\Servers\_120972e1d4ff4c498906e50d3499ed8c\arma3server_x64_2024-11-17_23-34-47_BegEndContext.mdmp
 2:56:34 Error: Mismatched Begin/EndContexts
 2:56:34 Error: Mismatched Begin/EndContexts
 2:56:34 Error in expression < > 0.9} || {([objNull, "VIEW", objNull] checkVisibility [_eyePos, eyePos _enemy]>
 2:56:34   Error position: <checkVisibility [_eyePos, eyePos _enemy]>
 2:56:34   Error CSIEW:GNED stack size violation
 2:56:34 File /z/lambs/addons/main/functions/UnitAction/fnc_doFleeing.sqf..., line 69
 2:56:34 Error: Mismatched Begin/EndContexts
 2:56:34 Error: Mismatched Begin/EndContexts
 2:56:34 Error: Mismatched Begin/EndContexts
 2:56:34 Error in expression <n_currentTask", "Fleeing (enemy near)", lambs_main_debug_functions];
    _unit s>
 2:56:34   Error position: <lambs_main_debug_functions];
    _unit s>
 2:56:34   Error Local variable in global space
 2:56:34 File /z/lambs/addons/main/functions/UnitAction/fnc_doFleeing.sqf..., line 73
 2:56:34 Error: Mismatched Begin/EndContexts
 2:56:34 Error: Mismatched Begin/EndContexts
 2:56:34 Error: Mismatched Begin/EndContexts
 2:56:34 Error: Mismatched Begin/EndContexts
 2:56:34 Error: Mismatched Begin/EndContexts
 2:56:34 Error: Mismatched Begin/EndContexts
 2:56:34 Client: Object 37:185 (type Type_132) not found.
 2:56:34 Setting invalid pitch 0.0000 for O Bravo 4-5:1 REMOTE
=======================================================
-------------------------------------------------------
Exception code: C0000005 ACCESS_VIOLATION at FBD66E76```

[whole cloud]

That was indeed already fixed, in 418, the last update

[broken phoenix]

Correct me if I am wrong as I am not the one who normally handles the arma updates on our server but this is the v9 build?
Version: 2.18.152413

[inland dew]

418 is v9

[broken phoenix]

Yikes, i have someone to yell at then for telling me we are on v9 😅 .

[ashen helm]

do you mean you want my profile files? You open the game difficulty, change Vision Aid to disabled, click OK and once you go back it's enabled again

[whole cloud]

413 was also v9, before we pushed a hotfix due to crashes replacing the first v9 release

[whole cloud]

But you said new profile fixed it, so its probably something specific to the profile files?
If its so easy to repro, then just ticket

[broken phoenix]

Ok, thanks for the help Ill get us on the hotfix so this will stop👍 .

[molten berry]

hey i can confirm this issue but with Crosshair

i think there might be something with difficulty settings not applying immediately

[empty goblet]

so guess i'm the only one with one 418 server crash so far

[molten berry]

because after i restarted the game the setting was changed, but before restart it reverted as if it didnt save the setting

[runic sigil]

Argh
I overthought where to put it
I even started in the right channel

[ashen helm]

do you happen to still have with this bug so you or I could send the profile files to Dedmen

I apparently fixed mine through chhanging difficulty to one of the presets and then making custom stuff again

[molten berry]

ill check

[lethal forum]

will it ever be?

[silk pewter]

most likely not

[knotty wraith]

then take off your label in Bohemia

the list is big Bohemia.net and there is no use from you

[silk pewter]

that's nice honey, how was school - were you bullied again?

[knotty wraith]

Uncle - I am very old

[silk pewter]

and I am BI, curious

[fickle geyser]

it's time for pills grandpa.

[knotty wraith]

If you can't fix anything, why are you writing?

[analog acorn]

Lou works on the documentation, not the code. His whole job is to write.

[knotty wraith]

fix the list bohemia.net and give those who have the opportunity to edit

otherwise it's just populism

[whole cloud]

no

[whole cloud]

I enthusiastically invite you to leave this channel. We don't need people like that in here. Thank you.

[fickle geyser]

changing FOV can break too many UI related things unfortunately

[median belfry]

get well soon!

[silk pewter]

soon you will ask me to apologise for being on the company game's Discord
take a seat, have a drink, kick back, relax, because we're not enemies here - unless you make it that way

[vivid reef]

https://media.discordapp.net/attachments/481382117910249483/1308111853276368916/image.png?ex=673cc199&is=673b7019&hm=6635ce5cac66cb902b5e473e5e5e88ca6cb1a65901652c70f8fd997efda0e1bd&
i tried the normal profiling and the profiling exe, i did expect a bit more of a difference

[ashen helm]

Why would they be different?

[vivid reef]

dont they have different optimisations?

[heavy vortex]

No, profiling just has a tiny bit of overhead.

[fickle geyser]

10 fps is a lot tho

[vivid reef]

3 fps

[fickle geyser]

and you have a lot of FPSs already

[vivid reef]

138 is stable

[fickle geyser]

yeah, misread

performance and profiling have the same optimizations

profiling has a bit of an overhead... as it has instrumentation allowing to... profile the performance.

so it should be tiny bit less performant.

they're the same source code built with different flags.

[vivid reef]

so its fine what my results were

[heavy vortex]

I'd have hoped for more stable -> perf difference, but I'm not sure which of the optimizations are active atm.

Can you check the difference in CPU power consumption over the benchmark?

[vivid reef]

the stable benchmark i did with 138 is some time ago already

[vivid reef]

maybe later, im in a game currently

62W is what i have as max right now, might be from ingame, might be from yaab

but from what i see, it wont be much difference, im still below 30% usage

[fickle geyser]

what cpu?

[vivid reef]

7800x3d, 6000mhz cl30, 4070ti

[fickle geyser]

nice, damn my 5900X pulling ~120+W and getting way less FPS ;D

[vivid reef]

its no secret the x3d perform very good in arma

im thinking about swapping it to 9800x3d if i can get hands on without a scalped price

[fickle geyser]

I'm trying to retain myself from upgrading to it as frankly, everything but arma runs great/good enough and I'm thinking about holding to next gen CPUs.

[vivid reef]

my 7800 is just 1 year old, so 50% of the upgrade is paid

[fickle geyser]

And the great work being done on Perf should make this CPU bit more viable for Arma, I need to find time to run some YAAB to see the difference.

[inland dew]

more stuff multi-threaded is always better, but still need best possible single core perf

[knotty wraith]

don't be offended, I love everyone

[vivid reef]

but some good thing, didnt had a crash since the last profiling update from dedmen

[silk pewter]

. . . b-but I also write code for Reforger's environment generators and wiki extracts 😅 😄
I just don't code for A3 though, true 😉

[analog acorn]

Sorry, my mistake. I don't keep up with Reforger as closely so I missed that.

[silk pewter]

no worries, just kidding around! 🍻

[vivid reef]

what exactly u want to see?

[heavy vortex]

Just looking for a rough estimate really.

[vivid reef]

so a yaab with a power graph below?

[heavy vortex]

Well, that would be ideal. Would need a comparison to stable though.

[vivid reef]

ez

https://media.discordapp.net/attachments/933089236654129202/1308136604015005856/image.png?ex=673cd8a6&is=673b8726&hm=eb5785b436511070c9aec2b4b8d24f97c703c68ac1883b9db4028e83a9f42e62&=&format=webp&quality=lossless&width=989&height=655 thats profiling with my usual game graphic settings, not exceeding 50W

https://media.discordapp.net/attachments/933089236654129202/1308137785680728174/image.png?ex=673cd9c0&is=673b8840&hm=d8d1ff1897c59cee4225e07972dbc9b008909cf71180b9296f572433c0b9eb01&=&format=webp&quality=lossless&width=989&height=657 stable, same settings

no diff in package power

[silk summit]

Profiling has the same optimizations found on the dev branch?

[fickle geyser]

dev branch is usually a bit behind, it's not updated so frequently.

[feral harness]

dev might have all the code that prof has and more but is not optimised at all

[restive pilot]

According to my test they perform the same (at least in YAAB/SP)

[loud wind]

Steam profiling branch (v09?) + Ampersand AVX512 malloc (https://github.com/GoldJohnKing/oneTBB/releases/tag/Arma-3-v2021.5.0-20211230) together gives me around 24% FPS boost on my 5800X3D

[restive pilot]

5800X3D doesn't even support AVX512
So it most likely doesn't do anything (just enabling AVX512 support by complier doesn't mean it'll generate code for it)

[loud wind]

That's weird - I thought it had but you're right. Oddly enough it still gives me a performance boost, running just the malloc alone gives a 10% perf increase over standard, more than the others I have tried

[eternal kraken]

i get the same performance improvements with this malloc like with mimalloc_lock_pages but with the GJK_Ampersand_malloc i have less RAM usage (mimalloc_lock_pages 20GB, GJK_Ampersand_malloc 16GB) -> https://steamuserimages-a.akamaihd.net/ugc/32182625622947600/455C7028E9631196ABB1E4286E307F96BB647FEF/

[loud wind]

mimalloc didn't run as well for me, but I didn't try the lockpages version. All of GJK's malloc builds seem to run really well over standard, and on top of the perf branch enhancements I'm still getting surprised by this 11 year old game 😄

[spiral pond]

did you try to capture the two dips yet with sframe cheat? (might not be low enough)
otherwise needs a mod to expose console in SP + diag_captureSlowFrame ["total", 0.025, 0, true, 100]; (below 40 fps + logToFile) [false for captureUI]

[eternal kraken]

not yet kju, but i will and report you

[tawdry gazelle]

True. mimalloc-lock-pages will reserve and allocate continues 1GB pages. I suggest not using this unless it works really much more better than others.

[eternal kraken]

i have 48GB DDR4 RAM and its the only malloc of the package which gives me an improvement, so no problem for me. but with the GJK_Ampersand_malloc i get the same improvements with less RAM usage. both mallocs (lock_pages & GJK_Ampersand) working for me. so thank you GJK for this 🙂

[whole cloud]

dev branch is built the same as performance exe on profiling branch. But with increased version number, and having the changes that are not prof compatible

[misty flax]

Guys, I have a question. After switching to profiling version on steam, I noticed I had both the arma3_x64.exe and arma3profiling_x64.exe binaries.

After launching the game through the a3 launcher, task manager said the process was arma3_x64.exe, not the profiling one.

So my question is, when switching to profiling, is arma3_x64.exe the performance build or is it the old one?

Do I need to replace the arma3_x64.exe with the profiling one by simply renaming the binaries to make the arma3 launcher use the profiling binary?

Or maybe I am just stupid and profiling!=performance, right?

and the performance binary is just the arma3_x64.exe

[analog acorn]

When on perf/prof branch, the default exe is perf, and the other one is prof.
They both have the same experimental changes, prof just has extra tools for logging and debugging, so if you don't need those tools you don't need to change anything

[spice dirge]

(said extra tools apparently making perfprof binary a tad bit slower as well)

[whole cloud]

prof* slower.

[spice dirge]

fixeded

[whole cloud]

arma3_x64.exe is basically "beta" version, having all the improvements that you came here for.
And "arma3profiling_x64.exe" is "beta" version WITH profiling on top. Making it a bit slower but giving you captureFrame tool.

📓 Couple finds from today

• Intercept causes crashes when running some script commands from a worker thread
• I broke the AI time limiting.

Old: ```sqf
_endTime = time + 7ms;
{
if (time > _endTime ) break;
simulate _x;
} forEach allGroups;

New: ```sqf
_endTime = time + 7ms;
{
  if (time > _endTime ) break;
  _groupsToSim pushBack _x;
} forEach allGroups;

{ simulatePart1 _x; } forEach _groupsToSim;
{ simulatePart2 _x; } forEach _groupsToSim;
{ simulatePart3 _x; } forEach _groupsToSim;
{ simulatePart4 _x; } forEach _groupsToSim;

Previously it would not start new AI groups after 7ms (Didn't help you much if one AI group was very heavy and exceeded the time limit anyway)
But with the new code, AI might even run to 20ms and still start new groups then, causing lag spikes.

Even after the fix it will be a bit worse because after groups have entered the first part of their simulation, they can't be stopped anymore. So if all groups start up, but then take so long that the time limit elapses, I still need to finish the groups that I've started. But they have multithreading in there, so they run a bit faster now so.. hopefully going to be good enough 🤞

Both will be fixed in v11

[eternal kraken]

i found a solution, i extracted the YAAB pbo, created a trigger in the INTRO part(where the yaab benchmark stuff is running) of the mission with diag_captureSlowFrame ["total", 0.08, 0, false, 5] and made a new pbo. Now i have my diag_captureSlowFrame in YAAB and here you get the Capture 🙂

[upper sierra]

That's pretty insane FPS, what kind of rig you got?

[vivid reef]

click my profile, its in my steam

[upper sierra]

Ahhh you have a 3D CPU damn, I guess it really does add a lot to FPS in a game like ArmA

[misty flax]

Ok thank you for the clarification

[feral harness]

Even from early days, dev was not performing as good as retail. Then recently I was comparing the improvements against legacy version so I had to generate retail exe because the legacy retail was beating dev hands down

[whole cloud]

dev is a retail build though, it shouldn't have any difference

[whole cloud]

And another bad one.
The AI sensors updating now has the time limit as it should have.
But, the time limit can be negative (apparently..) which makes it process zero sensors, so if FPS is low it will not process them, and keep piling up work, and the processing it does even if it doesn't process any, is probably going to keep FPS low still, that could be causing issues on servers.
Maybe thats related to the HC issues that were reported

[feral harness]

thought it was release built

[whole cloud]

that is the diag exe

[feral harness]

i was testing on release

[whole cloud]

2.18.152424 new PROFILING branch with PERFORMANCE binaries, v10, server and client, windows 64-bit, linux server 64-bit
- Added: Uniform textures are now stored on the uniform entity and persist in MP
- Added: FPS limit in the main menu with -world=empty and when windowed mode is unfocused
- Changed: Removed a few inventory related JIP messages (should reduce size of the JIP queue)
- Fixed: JIP players seeing a different uniform on a player for up to a minute (please test looting from dead bodies as this might have indirectly fixed a few bugs)
- Fixed: Unit insignia were sometimes being applied to the whole body - https://feedback.bistudio.com/T155459
- Fixed: Several crashes

If you don't want to use the Steam branch, the files are also available for alternative download here:
https://drive.google.com/drive/folders/15p9j7C2nHUt6NoVfChX4YFuqzFXzblJh

[whole cloud]

Most of these features require perf branch on server side

[vivid reef]

read changelog
read dedmen additional info

[knotty wraith]

https://youtu.be/PtmZf4WoaLA
v10 - HC AI

[runic sigil]

Are more YAAB benchmarks with other hardware combinations useful?
Prof vs Stable was what I was thinking

[whole cloud]

looks like big lag spikes
Hard to see with #monitor so slow that it averages them out. Also #monitor is only the server, while the problem is on HC.
Most useful would be a capture slow frame from the HC. its probably running very low fps

[knotty wraith]

I understand, but I had to roll back to the stable version - it’s impossible to play. I'll try to do a more detailed analysis one of these days.

https://youtu.be/qaiGNsxyD5c
stable - for comparison

[spiral pond]

next time use #monitor 1 and [diag_fps,diag_deltaTime,diag_fpsMin] for the HC (via CBA console with variable watching on remote HC)

[knotty wraith]

CBA console?

[spiral pond]

one moment

[vivid reef]

crashdump in your dm

[knotty wraith]

I use this for myself

0 spawn {
ddd = true;
VAR_FPS_HC = 0;
_code = { 
 VAR_FPS_HC = diag_fps;  
 publicvariableserver "VAR_FPS_HC";   
};
[_code] remoteExec ["call", HC];
//VAR_FPS_HC

while{ddd}do{
[_code] remoteExec ["call", HC];
sleep 0.7;
(str [round VAR_FPS_HC,round diag_fps, count allunits, count (vehicles select {count crew _x > 0})]) remoteExec ["systemchat", -2]

}}```

[spiral pond]

as for slow capture: #captureSlowFrame total 0 (only for the DS itself) and for HC: diag_captureSlowFrame ["total", 0, 0, true]; executed via CBA console on HC for a single capture at this moment, or diag_captureSlowFrame ["sLoop", 0.1, 0, true, 10]; (to capture 10 situations with < 10 fps)

(means you need to load CBA on DS+HC+client, allow via bikey or disable signature check - plus allow in missions via https://community.bistudio.com/wiki/Description.ext#enableDebugConsole or load another mod alongside that enables console for all missions)

[carmine stump]

not important, but in beta tab profiling should be one option and performance another. I see a lot of people get confused (as I once was) and I guess most just want the performance exe, not profiling.

[whole cloud]

Yes. Most just want performance, that's why everyone gets it as default and only the professionals needing profiling have to manually switch to it

[empty goblet]

hmm, does this thave the missing scope fixes for captureFrame or must wait for v11 ?

[whole cloud]

v11

but seems v11 is tomorrow already

[empty goblet]

to wait or not wait for servers 🤣 that's the question

[whole cloud]

Working on better JIP logging.
Server savegames, actually store the whole jip queue in the savegame file.
Looking at that, Lets say, which inventory item was removed from a unit:

[empty goblet]

so tried to join with 424 to 418 server and got instant client crash hmm (after role selection), replicated 2nd time

[whole cloud]

or public variables?
But you cannot see the contents of variables because its encoded there.

[empty goblet]

dumps sent toward dedmen

so i guess i'm not risking servers with v10 😦

[whole cloud]

ugh known crash in the uniform texture stuf

[pure dove]

Hello, is there any reason why freeExtension command is only available in dev branch? While I often switch to dev branch for extra debug info, I just got curious

[whole cloud]

in every way unrelated to #perf_prof_branch
Because unsafe and security

[pure dove]

oh sorry about that, thanks anyway

[vivid rune]

V10: Animation stutter in the beginning. After some time whole Arma freezes in "Eastwind Campaign/Signal Lost" still present.

[empty goblet]

v11 can't come fast enough 😁

[blissful pilot]

hello, got told to come here
#arma3_troubleshooting message

is the profiling build causing crashes?

[whole cloud]

yes, but without mdmp unsure if its the same

Probably switch away from #perf_prof_branch
fix coming tomorrow

[blissful pilot]

can i send one here?

[whole cloud]

or my DMs

[light geyser]

Is anyone experiencing STATUS_APPLICATION_HANG error when you run Dynamic Recon Ops or other scenario?

[whole cloud]

⚠️ We know about crashes on prof. For technical reasons we cannot fix it until tomorrow.
Consider switching away from #perf_prof_branch 🫂

[indigo anvil]

😦

[empty goblet]

or manually download v9 client from google drive

[short hollow]

To clarify, performance and profiling branches are essentially the same and casual users should default to performance?

[ashen helm]

casual user will download both after choosing the profiling beta in Steam and performance exe will be used by default

[empty goblet]

performance binary is the default in profiling branch, the profiling binary is just more diagnostic stuff (lower fps)

[short hollow]

I see, thank you

[naive osprey]

Does the uniform texture persistence also include backpack textures?
(backpackContainer player) setObjectTexture[0, ""]

[silk pewter]

isn't that already the case? I remember CTFs using backpacks as flags 🤔

oh, but it may be lost if put in a container…

[naive osprey]

Potentially? We had to make a workaround for a similar issue in 2021. It included uniform + backpack textures not persisting if the person is well outside the render distance. Maybe it was fixed previously, https://feedback.bistudio.com/T157449

If I could get rid of our workaround, I will die of happiness

[rapid wing]

keep crashing? is that something to do with prof branch ?

[dreamy yacht]

^

[harsh grotto]

download v9 or install main branch

https://drive.google.com/drive/folders/15p9j7C2nHUt6NoVfChX4YFuqzFXzblJh

[feral harness]

backpack has own textures you set on backpack for a long time and it is persistent. uniform texture was stored on player, now it is stored on uniform so you can set texture on it using uniformContainer as well as player

Also you need to use setObjectTextureGLOBAL if you want persistence

[naive osprey]

Was just using an example, wasnt the exact script.

[spiral thicket]

Yo! I noticed on performance branch that Arma would crash whenever trying to load certain textures (for me: any RHS ch47 / TF373 chinook, modded osprey)

[naive osprey]

#perf_prof_branch message

[spiral thicket]

Yeah I saw just naming some of the causes incase it helps correct it

[feral harness]

yes thats the reason for the crash dedmen explained

[whole cloud]

2.18.152431 new PROFILING branch with PERFORMANCE binaries, v11, server and client, windows 64-bit, linux server 64-bit
- Added: JIP queue export now also exports the queue's contents (Profiling binary with -debug parameter only)
- Changed: JIP queue export is now multithreaded and freezes the server for less time
- Fixed: AI simulation would not honor its time limit
- Fixed: AI visibility updates could get stuck not processing any updates
- Fixed: Crash when script commands are executed from extension threads
- Fixed: Crash when joining servers
- Fixed: Keybinding F16-F23 keys were displaying "UNEXPECTED_KEY_ID" instead of the key name - https://feedback.bistudio.com/T125049

If you don't want to use the Steam branch, the files are also available for alternative download here:
https://drive.google.com/drive/folders/15p9j7C2nHUt6NoVfChX4YFuqzFXzblJh

[spiral pond]

diag_captureFrame 0; doesnt work - only starting with 1. I think to recall this was different earlier, was it not?

[whole cloud]

I think previously 0 would work same as 1.
It cannot capture the current frame, it can only capture the next frame or later. So the minimum that works is 1

Mh looking at code, I don't think 0 worked before

[light geyser]

The game still crashes when i open Dynamic Recon Ops

[whole cloud]

Send me crash report please

[light geyser]

Arma 3 Troubleshooting Report

Sorry for the inconvenience.
Please report this on Arma 3 Feedback Tracker (http://feedback.arma3.com). Thanks to you we may fix the issue faster.

---

Process exit

Exit code: 0xCFFFFFFF - STATUS_APPLICATION_HANG
Running time: 00:08:20.5196526

---

System information

Date: 2024-11-20 11:50:52 / 2024-11-20 12:50:52 +01
Current dir: C:\Program Files (x86)\Steam\steamapps\common\Arma 3
Command line: "C:\Program Files (x86)\Steam\steamapps\common\Arma 3\arma3launcher.exe"
CLR version: 4.0.30319.42000
OS Version: 10.0.22631.0 (Microsoft Windows NT 10.0.22631.0)
OS 64 bit: True
Process 64 bit: False
Culture: es-ES
UI Culture: es-ES

---

Last game report

not provided

[whole cloud]

I mean the report file, that the "Prepare report" button exports, the zip
This text tells me nothing besides that the game actually didn't crash but just hung.

[feral harness]

Multiplayer?

[light geyser]

No, singleplayer

[feral harness]

try creating a new soldier profile

[vivid bear]

Is the units flickering issue now fixed?

[feral harness]

I just downloaded this scenario for Stratis, no hang or crash for me

[light geyser]

It freezes and crashes during the "Faction extraction data"

[feral harness]

Tried new profile, also no problem but some script errors

[feral harness]

with new soldier profile?

are you on latest prof?

does it do this on current stable?

[light geyser]

You mean a carpet named "ArmaReport_Log_20241120T115052_" or something?

[whole cloud]

yes

[light geyser]

No

[light geyser]

I don't know how i'll can send you, but i'll send you the Google Drive link

[light geyser]

https://drive.google.com/file/d/1YXwRF7Y10mwU-eT8l6pzZ3tyh7Af17_1/view

[whole cloud]

You also need to give everyone access to view it

[light geyser]

one moment

[light geyser]

https://drive.google.com/file/d/1YXwRF7Y10mwU-eT8l6pzZ3tyh7Af17_1/view?usp=sharing

[light geyser]

I'm afraid of losing all loadouts and composition i created

[feral harness]

I can see you have a million of mods and custom start up params

[light geyser]

Is that a problem?

[feral harness]

Could be in this case

Does this cocktail work well with stable?

[light geyser]

I used the same mods before and the DRO didn't crash

[whole cloud]

Please get rid of these start parameters, they are nonsense
-cpuCount=7 -exThreads=7 -enableHT -malloc=tbb4malloc_bi_x64 -maxvram=2817 -noLand
And I don't think you want to have that maxmem one either. Unless you specifically want to limit the game

[light geyser]

Which file?

[whole cloud]

Start parameters, in the launcher

Its a bug in prof indeed.
Switch away from profiling branch for now

[light geyser]

What does it mean enableHT ?

[whole cloud]

In your case, it does nothing because you set -cpuCount. So its just completely ignored.
If you don't know what it means, that means you shouldn't use it

[light geyser]

Sorry, my parameters are in Spanish, so i have some problems on find it to get rid of it 😅

[whole cloud]

I don't know spanish.
Just look through all things that are checked.
I would say just uncheck literally all of them

[heavy galleon]

btw for server, should I -enableHT or not? Two people told me two different things

[whole cloud]

Test what performs better

[heavy galleon]

how would I test server? Just spam fuckton of AI?

[whole cloud]

Start some benchmark scene, remove fps limit

[heavy galleon]

where yaab for servers?

[silk pewter]

isn't it enabled by default? also, -enableHT is… an old setting that is good in autodetect, no?

[whole cloud]

enableHT is not enabled by default

[heavy galleon]

so, test -enableHT (will find a way I guess), stay away from -cpuCount all together and leave that for autodetect

(even on linux)

[light geyser]

I got rid of those parameters and i still have the same problems when in run DRO. It freezes during loading Extracting Faction Data 😖

[inland dew]

always -enableHT, unless you use -cpuCount

[whole cloud]

As I said above, #perf_prof_branch message

[inland dew]

unless HT/SMT is for some reason disabled in the bios

[heavy galleon]

and switch off from profiling for stability this Sunday

I assume there`s no “easy” way to simulate large pvp on a server, right?
AIs have fully different processing right?

[empty goblet]

ok, something is wrong with v11

client frozen, server frozen

[whole cloud]

#perf_prof_branch message

[regal lance]

Alo! I don`t want to disturb any of you, especially Dedmen. But the profiling branch is not working to us. It opens the Arma3 "terminal" but it gets stuck on Reading mission...
Rolled back to the stable build and working as intended. I can find those rtp if you want.

[whole cloud]

#perf_prof_branch message

[regal lance]

Dumb of me, I`ve should just read the chat before writting

https://tenor.com/view/bai-see-you-later-bye-adios-got-to-go-gif-21054483

[empty goblet]

sad noises ... waiting for v12

[whole cloud]

v11.1

[silk pewter]

talking about DirectX?

[light geyser]

I got rid of those parameters and switched away from profiling branch, now the game works when i run DRO on singleplayer. Thanks

[empty goblet]

so sneaky jab ...

[deft pendant]

yo anyone know how to fix this memory crashing problem am tryna host a server on antistasi or even in the vanilla game and aparrently its out of memory i

i was playing it couple months ago and it worked

[silk pewter]

Hiyo! #arma3_troubleshooting can help you 😊

[heavy galleon]

Vulcan confirmed?

[midnight oyster]

Anything to get rid of my VRAM memory leak.... please

[ashen helm]

just get more VRAM

[silk pewter]

"just when it is about to reach the limit, download more VRAM"

[molten berry]

https://tenor.com/bg4nl.gif

Me when dedmen doesnt fix a crash 10 minutes after it occurs

[opal hound]

me when the bleeding edge branch is bleeding edge

[empty goblet]

⏲️ ⌛ ⏳

[grim oriole]

https://tenor.com/view/soon-tm-soon-space-gif-23401891

[whole cloud]

The freeze made me go back to 2008 💀

about 90 minutes out, hopefully

Or like 3 hours, depending on how the builder feels today

I couldn't reproduce the freeze, even running DRO but the wrong thingy is fixed

[heavy galleon]

Send code, I'll build locally

[harsh grotto]

https://tenor.com/view/cat-wait-waiting-cat-wait-waiting-cat-waiting-gif-9780709586447195996

[glass lagoon]

with proton steam's selection of performance_profile was crashing on joining mp on official servers, while dev was hanging on start showing arma3 logo - same settings, also under same exact settings performance profile yields less fps for me

the vram limit setting tooltip on max 8192Mb is maximum the game can utilize?

[molten berry]

Yeah I think 8gb is an engine limitation. Remember the game came out 11 years ago when most gpus had like 2gb of vram

[glass lagoon]

so in this sense it received no attention, is that what you're saying?

[whole cloud]

There is no vram limit.

The start parameter doesn't work

[molten berry]

lol thats awesome

[glass lagoon]

then why it is in the launcher and why doesn't it consume whole 20Gb of my vram presenting me with 30-60 fps

should i set ultra to offload more to gpu?

[whole cloud]

Because it used to be a parameter a decade ago, and no-one removed it

[glass lagoon]

oh

[whole cloud]

It doesn't use more because it doesn't need more

[glass lagoon]

well it clearly does since bushes don't appear instantly when i ride my offroad really fast

or am i oversimplifying it? kek

[whole cloud]

How fast bushes appear depends on your hard drive speed, not on how much memory your gpu has

[glass lagoon]

my hard drive read speed is 5Gb/s

[whole cloud]

That's not how hard drives work

[glass lagoon]

yes, any difference to nvme in context of what you're saying?

you know it just hangs on direct pci lane with underutilized cpu

bet it is dangling doing almost nothing to utilize its fancy iops rw capacity

[molten berry]

bro stop throwing out fancy words like you know what all of it means while still completely missing the point

[glass lagoon]

i don't

[molten berry]

let me go with the analogy of internet. You have latency and you have flow.

[glass lagoon]

i play it for 4 days and i had 4 encounters of simultaneous kills with hand held auto weapons with enemies on official mp servers

ping <60

[molten berry]

your "ping" could be really high like 200ms while still being able to download 5gb/s

[glass lagoon]

then shouldn't the in-game metrics be corrected?

allusions are not facts

[molten berry]

its a similar idea here with the disk loading. You need a lower latency disk in order for stuff to be loaded faster. It's why SSDs are better because a HDD needs to physically spin up and achieve the correct rpm and everything to start reading/writing. I might be oversimplyfing this explanation but the analogy to internet speed is the simplest i can think of

cause the game when youre moving is constantly swapping levels of detail of objects

[glass lagoon]

alright i see that this is not an applicable issue then

i don't have swap instead i got 4 banks of ram 32Gb each in 3600xmp

dual

[molten berry]

[spiral pond]

#360648487378878464 👮 needed

[glass lagoon]

why? could you please share the fps difference with stable vs performance profiling branches?

guys, did you know that linux kernel has more than 100k metrics for performance inside it? coincidentally there is a perf tool to read it out

have anyone tried that with proton?

how the "profiling" is happening

[burnt tartan]

https://downloadmoreram.com/

[eternal kraken]

i have a question about slowFrames and FPS. i observed that it happen that i achieve diag_fps > 50 but catching a slowFrame with diag_captureSlowFrame ["total", 0.06] (>60ms). vice versa can i get diag_fps < 38 but using diag_captureFrame 1 in this moment the MainThread totaltime is < 25ms or even if i run diag_captureSlowFrame ["total", 0.04] it is not triggered in this scenario with diag_fps < 38 (so the slowFrame/mainThread total time never last longer then 40ms). How are FPS and slowFrame connected to each other and how can it be that the Thrdad totaltime is < 40 ms but the FPS < 38.

[vivid rune]

v11 has still animation stutter and Arma freezes in the Eastwind campaign
#perf_prof_branch message

[whole cloud]

totaltime is < 40 ms but the FPS < 38
38 fps are 26ms
30fps are 33.3ms

So.. Thats how it can be that totalTime is less than 40ms and fps is less than 38

[eternal kraken]

what is the formula to calculate this, i didn't undertand it fully yet

[whole cloud]

frames per second.
Into time per frame

1 second / number of frames == time per frame

[empty goblet]

you do realize that's how physics work , if two shooter push trigger at same / similar moment with theirs weapons facing each other ... well ... there is two less shooters in the world

[quaint flame]

1000/fps

= ms per frame