never seen that on windows server 2012 r2 64bit running 1.64 for months
#perf_prof_branch
shell mountain
nocturne obsidian
Thats what our server was running @shell mountain
Windows Server 2012 R2 64bit
@timid coyote This is related the the headless client I am not sure why.
Because we also had a headless client. That was connected at the time.
shell mountain
@nocturne obsidian are you talking about latest 1.64 stable on win2012r2 ? meaning (1.64.138732) ?
timid coyote
we also have headless clients
@empty goblet is there anything known about this? is it maybe fixed in current perf?
empty goblet
@timid coyote just use the new perf , there're too many fixes
timid coyote
do i need the jemalloc in the dropbox?
empty goblet
not unless you want to use it, default is tbb4 (check v12 and the latest TBB4 is there)
timid coyote
what would you recommend? tbb or je?
shell mountain
hey guys, i tried to put together some basic fps trend stats from yesterday unpatched and today with v14:
http://imgur.com/a/TrM8K
after a good start unfortunately the fps drop behaviour seems still there
sry shitty pics and notes but i think is visible
i'm re-enabling my logging and visualization framework soon to monitor better this behaviour
i had quite some hope in perf build but it seems i still get serverfps locked to 5-7fps after a while,
do you also have this beahviour with perf build too on linux or windows?
carmine stump
@timid coyote I recommend giving jemalloc a try.
@empty goblet there's something wrong since V5 that are related to fps drop, you gotta check what was changed there.
amber totem
@empty goblet I am still getting the 5 FPS bug my friend.
Overall FPS went up, but we still get that bug D:
amber totem
@empty goblet I PM'd you.
amber totem
@Nostromo#4983 Should I revert to V4? Do you think that will fix my issues?
I was originally running V9 I believe
empty goblet
v4 is pre-hotfix #2 so obsolete ... also those 'something isn't right' are usually just placebo effect ...
carmine stump
@amber totem Im not running perf since v5, because I get the same 5fps issue as you
amber totem
So you downgraded to V4 and no issues?
amber totem
That was what I was originally on and had the issue
So, I am gunna give V4 a try, I suppose.
amber totem
My hypothesis is this - Other Life servers don't get this becuase when they opened, these newer branches weren't out. When we opened, we opened on a later branch. As most of these life server network managers don't bother updating, they are on the old ones.. thus, they do not get the issue1
I am just saying Life server because I host one
empty goblet
the low fps issue was there before so it's not v4 or v5 relevant ... it's just when people run out of memory
if there was mod update it could just push it beyond the treshold
shell mountain
got a better representation now, collected data of today with arma3server v14 on linux
http://imgur.com/a/sR4r5
I believe that serverfps deterioriation is quite clear from the picture,
The 50fps peaks are the server restarts and then the down slope after running over time (starting from ~20-30min on)
carmine stump
@empty goblet it's not related to memory leak, I get the issue the second I log into a server
shut mason
Not playing A3. It's a regular issue.
spiral pond
BI getting serious about fixing it
However it seems to me that they want to hold out until the 64bit build and hope that this will help "somehow"
nocturne obsidian
BI did the same with DX12 over 1.5 years ago. Its always been this way, performance isn't going to change unless BI genuinely commits to that goal, which they haven't. They have infact said multiple times not to expect any big performance change over the life time of the game.
empty goblet
@nocturne obsidian DX12? where ?
nocturne obsidian
Sitrep 139 I believe is when you first told people you were working on it. You didn't say it would improve performance, but just with 64 bit people are speculating wildly just as they did when DirectX 12 was mentioned. As I say you have never committed to fixing the games performance issues. BI just name drops things periodically that people associated with performance improvement and then in the vacuum of no further update they get excited. The parallels between the 64 bit and DX12 one are uncanny
Its more a criticism of people and hype culture based on nothing rather than BI really when you think about it.
empty goblet
@nocturne obsidian my point was it's not out ... it was said we looking into DX12 and that's it all ...
spiral pond
64 bit will come and it will improve the situation for some aspects. so i cant see a similarity to the DX12 investigations BI did - it was fairly obvious that it wouldnt help (much) and the effort being unreasonable for the limited gain (if any)
still its disappointing that BI seems to have given up on fixing the memory leak/memory crash/3fps bug after the last/the patch before
empty goblet
64bit main purpose are memory ranges, any other gains (from native 64bit code and memory blocks on 64bit OS on 64bit CPU) are minor (fraction of % to <3% depending on type)
spiral pond
yep
empty goblet
the main problem is that not enough people use profiling binaries to catch the memory leak ...
note: it's not in performance binaries ! - it needs the debug layer on DEVELOPMENT/PROFILING binaries โ
spiral pond
what about forcing everyone to use it for a week? ๐
i guess it needs a new thread with simple todos steps and SITREP or sth to make people aware of it
queen owl
How about first switching binaries around in the Steam PROF branch to make it even easier. Then a big noisy campaign on all fronts to convince people to enable it to get better performance possibly and help BI in the process.
swift field
What about trading promises of fixing particular bugs with some communities doing testing campaign with your PROF branch? ๐
nocturne obsidian
I'd run the profiling binaries
But I would just run a mission where AI vs AI happened
So an entirely client non-dependent mission.
@empty goblet So this is why everythings been so hard to find the memory leak?
No one wants to run the dev-branches?
thin wyvern
@empty goblet you need to pinn your last message because this is crucial information for the people who have 3 FPS bug.
low narwhal
I've seen that error message @timid coyote not seen it since we started using performance binary for the server, can't recall which version.
timid coyote
we're using perf now, too - so far no crash
shell mountain
with the perf branch the bug changed from 5fps fixed, to 7-9fps.
if helpful I will switch to Profiling branch to provide anything useful.
@thin wyvern honestly i also thought perf branch would have some profiling included.
@empty goblet is there anything else to be done beside switching to use the profiling binaries?
I mean reporting logs or dumps?
low narwhal
Any players who's not using mods getting the 3fps bug?
carmine stump
yes jager, I get low fps with perf branch, even with no mods
low narwhal
Cheers nostromo, been having some acre issues as well so I was half hoping that was related.
zenith swallow
Hey, does anyone know anything about a 5FPS bug occurring for people. I mean, for individual clients on a server? I've talked to a few people, and looked through logs and it all points towards SquadXMLs being broken. I've had issues with this in the past about servers crashing. But never like this, anyone know anything about this?
cursive vigil
@zenith swallow It's a widespread issue and devs are aware. It has nothing to do with squad.xml, it's a memory leak (or growth) issue, once Arma runs out of memory the FPS tanks
They've not yet found the cause, last word I saw on the subject was that not enough people affected were running the profiling binary (not the same as performance binary), so they had little debugging info to work with
and I guess they can't reproduce the issue themselves, otherwise they'd not need that user submitted data
low narwhal
So I ended up using profiling binary and managed to make the 3fps bug appear, however it does not seem to want to crash. Arma 3 just hangs it seems, not sure if it will produce anything of use if I just kill the process
Anyone know if I'll get a bidmp or a mdmp when it crashes?
eager pasture
A crash should output both files.
low narwhal
Yeah that's just the thing though, it's not "crashing" it simply goes black screen and the process hangs
It retains the memory usage it has had but next to nothing happens on the CPU side
If I kill the process via task manager will it generate the files or will it output nothing asides from the rpt?
eager pasture
Give it ago; kill the process.
low narwhal
Bah, that's nearly two hours down the drain, no crash dumps
Back to the drawing board, cheers anyways
eager pasture
๐ฎ ๐ข
low narwhal
Makes sense why it's hard to narrow down if no logs are generated.
nocturne obsidian
Would not a diag_captureFrame at the very least be useful in that circumstance? You could export the results of that and it ought to point out where the issue is occurring.
low narwhal
That sounds a tad above my expertise, is that something I could run through the console?
nocturne obsidian
On the client in the escape menu you just type diag_captureFrame 180; and in 3 seconds time or so (its 180 frames) the dialog will appear and its got an export button and a UI to look through the data. If you want to do it attached to the server you need to be logged in as admin to have that menu option to appear. Just run it locally, easy as that really.
queen owl
No need for the console or admin. Just hold left shift, tap numpad minus, release shift, type 'frame', and the capture dialog will open.
low narwhal
Cheers, I did get some attempts on capturing the bug but it's really too random to capture it seems and when it strikes there is limited time before the process dies entirely.
zenith swallow
...
spiral pond
low narwhal
After 92 minutes of koth I finally got the capture frame
nocturne obsidian
Nice. Now BI will do the same thing they did with my traces of the past, never mention them again (I wish I was joking but having got an entire 35 person community to profiler build for a complete Sunday of gaming and provided traces they went silent and stopped responding and no apparent progress either)
It sits there in the forums as a testiment of just how little they will do in response to the community helping them track their problems. I sure hope more comes from your efforts since it took quite a bit of time and effort.
carmine stump
it would be so much better if they actually said: ok, our engine is bad, there's no way to fix it, that's why we are already working on a new one to use in our next game.
empty goblet
@nocturne obsidian i'm quite sure i already thanked you for your effort and used some of your crashdumps / reports in past ...
nocturne obsidian
@empty goblet nope you have never thanked me for a single thing. You have dismissed 90% of the bugs I have reported off hand and no one ever responded to the profiler traces despite BI staff asking repeatedly for them to be done. It has never happened, not for our crashdumps or for the serious game breaking bugs we found the causes of not ever. No point starting now, its been 6 years of being treated poorly at this point we don't undo that abuse by being sorry not sorry now.
empty goblet
@nocturne obsidian on skype, we discussed last in 2013 , on discord i've no history with you, can't find anything meaningful in email, nor forums history (pms or posts), same goes about public channels ... anyway captureFrames were processed in the period when they were needed ... anyway i processing 100s of crashes, freezes, logs, performance, security or mp issues each major build and it's known i sometimes fail to thank for some because there is other work also ... also i will dismiss any nonsense or bogus stuff, that's known of me
spiral pond
so captureFrames are no longer needed despite the bug not fixed???
also Dwarden the criticism is very rarely about you - the others in the A3 team fall short..
for the most part there is no point in posting about issues aside from crashes and issues with the launcher (and to an extent scripting issues thanks to KK) - for these in the FT there is decent attention; for the rest is just totally random selection with very low chances to get attention
from the outside perspective
the whole feedback loop to the community would need a serious revision for a better approach, working tools and more resources, yet it seems low on the prio list ever since A3 (before it was only better thanks to Ondrej, a few other individuals, and you)
cursive vigil
@spiral pond When he said 'capture frames were processed in the period when they were needed' I think he meant for historical issues, not for the memleak issue
fwiw, backtraces/memory debugging don't always reveal the cause of the issue, in many cases they only point in the general direction of the problem or they just don't reveal anything useful at all, but you can't know until you've seen them. If they could reliably reproduce the issue I'd expect they'll have done a binary search of commits in the period to find the trigger
spiral pond
@cursive vigil you are mistaken - they did ask for frame captures for that issue
in any case i am a programmer myself; obviously its not simple, yet during A2 days I was lead of the community issue tracker team and we had a close cooperation with their lead programmer and Dwarden (and a few more)
sadly the A3 team didnt care about that all, dumped us and everyone and left things in shambles
I give them credit that they tried to rebuild it from scratch and did some things better too (broader focus on the A3 team at first for their issue tracker); however the overall effort was never near as well working as during A2-OA times
while i dont blame that much besides all other things, to have no issue tracker for like 6-9 months and not trying to address the issues with the new system, phabricator, just tells you how little importance they put into this
(and again Dwarden is exempt from the critism - the fault lies in the leadership responsble for that)
cursive vigil
@spiral pond He didn't say they no longer required captures for that issue?
which is what I was saying?
I know they asked for them for the memleak issue, if you scroll up just a little I even explained that to someone who was talking about the problem. However nowhere have I seen Dwarden say that they don't want or need them anymore for that bug.
empty goblet
@spiral pond i was talking in generic, there were several 'waves/events' when captureFrames were asked specifically (same goes for full memory dumps etc.)
cursive vigil
@empty goblet If it helps, it's periodically reproducible on Linux too with 1.58, in fact it happened to both myself and a couple of people last night after a couple of hours of playing the 'Liberation' Coop
spiral pond
@cursive vigil in the forum the lead QA guy that they stopped looking into it (with debug builds) and ppl no longer are to send them as a result
(also recently he said that they may look into it again)
still so widespread the issue is with such consequences it is very disappointing to not see more prio for it
cursive vigil
ok, I can't read everything posted on the forums, I thought you were referring to something Dwarden had said in here
spiral pond
well it just shows another problem in the communication here. i do applaud that they reach out via the news items to fans at times, also for this one, at the same time they dont keep people updated that way still (probably in fear of the negative backslash)
nocturne obsidian
There is a very real issue around the status of what is getting worked on especially for mod authors. On the one hand Sitreps and such a big improvement but they are also too high level to track a particular issue. Then you have the individual status of different areas (like AI) but that is disconnected from the release schedule and its not clear when any individual change is coming through. There is lots of "hidden" development like the audio update which left LordJarhead and JSRS just hanging there with 6 months of work destroyed just one week after release. We have had quite a few big mods just quit over the continuous pressure of staying current that dominates development. The updates we get aren't sufficient considering Arma is a platform on which the games people play are developed and based. We have neither the versioning system available that allows downstream to work at its chosen pace nor the changelog detail and migration information to keep current.
shell mountain
jfi: running latest profiling build on linux
queen owl
Wasn't the sound update announced like a year beforehand and was available for testing months before the stable update? Didn't seem like much of a surprise really
thin wyvern
@empty goblet Can you add fix from Dev/RC branch for squad.xml bug in next perf/prof branch https://feedback.bistudio.com/T120466
spiral pond
@BrightCandle#1562 LJ is not much into configs. it wasnt really that much of a deal. he makes drama at times
nocturne obsidian
@spiral pond I can see why he was annoyed. I can also see why multiple mod makers have given up over the years after the maintenance got crazy. We maintain a host of custom stuff and every release its a mad scramble to get it ready for the weekend game, or to tell the entire community to do a rollback. We have to make a snap decision about how broken the game is and whether the mods are broken or not. Every community using mods has the same problem and the more custom stuff you maintain the worse it gets. Its not the updates are an issue its the dependency and how its not in our control. I would very much like to go back to 1.58 frankly because while the visual update and such is worthwhile the broken AI (spotting distance especially) has broken our game for months at this point. Can't be done. Everyone is forced to follow along and that is bad for development, no one does it this way.
spiral pond
well its different problems. the steam system certainly has also its downsides to before
still also many modders are quite clueless - they are hobbyist. the main fault is the lack of docu and good tutorials by BI of course. still many just dont have a clue how things work and dont bother to learn
this is generally why teams are useful - to have different people with different skills. or at least cooperate with others. this does happen in general.
yet for those complaining in general doesnt happen (enough)
spiral pond
for your particular problem i guess you could use arma3sync and even put A3 on there (if the version is still available somewhere); with withSIX tools you can also configure precisely what mod versions you want and put A3 itself on your custom repo
in regards spotting distance/AI - its fairly simple to tweak via a config mod. it just needs considerable time tweaking and testing
Zeus community usually did that in the past and shared their version. most other AI mods usually change a lot more, with scripting, that is not always desired due multiple reasons
rapid fable
^ I second the need for proper documentation and in depth detailed tutorials.
nocturne obsidian
Its the move to steam that did it, its actually a very bad platform for modded games due to its lack of version dependency control. It isn't something I think BI can ever really fix, there are too many happy customers of Steam and it hurts the milsim niche for A3 it is what it is. We do what we can in the sense we have a 100% custom repository for mods and missions with swifty (which is far simpler for new users than arma3sync) and its about the best we can do currently.
@spiral pond can you point me to the appropriate bit of config that allows us to increase the maximum spot distance, just another default behaviour to change in a sea of stuff we have done in the past few years if that is all there is to it we will fix it.
empty goblet
v0 ready
1.66.139519 new PROFILING branch with PERFORMANCE binaries, v0, server and client, windows/linux
- sync with 1.66.stable (with 1.64.perf changes/features which didn't make it into 1.66.stable)
https://www.dropbox.com/sh/582opsto4mmr8d8/3BSy9PdRGm
https://drive.google.com/folderview?id=0B03-H4YIbhkFMUt5RzNqZjFlNGs
available via STEAMklient/STEAMcmd as branch too, read https://community.bistudio.com/wiki/Arma_3_Steam_Branches#Arma_3_Server
nocturne obsidian
Does this fix the problem with onmpkilled?
bright pond
+1
It doesn't seem so, on my exile server with v0 perf onmpkilled still doesn't execute.
I must add that I only tested with the build serverside and not clientside
rapid fable
It does not help that you're testing with nonvanilla Arma
It's like complaining to the Ford dealer that you're speedometer is still indicating the incorrect speed after a service, but you run oversize tires.
Mods are akin to "aftermarket" parts. They change how the base game plays. As such you will never get an accurate representation of vanilla performance and stability improvements.
bright pond
Yeah I understand that but my test was just to see if it worked like before the update...
I don't have the repro mission that was used here so can't really test https://feedback.bistudio.com/T121736
rapid fable
Unfortunately not many understand that. Perfect/Profs aren't meant to fix issues in nor improve custom Arma mods.
hallow yew
Most players buy Arma 3 to play mods though
rapid fable
I agree. Maybe a house analogy would be better. Arma is the foundation upon which a mod is built. If for some reason BI alter or move the foundation, everything built upon it must be altered. So, when people say "this improvement didn't help at all, in fact, it made something worse" they need to understand that the "improvement" wasn't meant for them in the first place but instead for the core game.
Some have been around long enough to see how cyclical this issue is.
One only has to look at how the Apex lighting improvements broke many terrains.
bright pond
The problem is when at least >70% of players play ArmA for the mod built on it the devs have to (if they want to please the majority of players) take in consideration the mods for the majority of things they do.
rapid fable
To a certain extent yes, but overall I'd have to disagree. Mods wane and grow with the wind. What's popular today may be gone tomorrow. You can't expect a company to base its business model on something so mercurial.
I have all the respect in the world for mod makers. But there is and has always been one rock solid rule. Any change in the core game may and will break your mod. That's a risk all mod makers accept when they begin.
hallow yew
Has there ever been an Arma update that didn't break something in a mod though? ๐
rapid fable
Not that I'm aware of, that's why I'm wondering "why all the fuss".
bright pond
First I have to say that from a business pov i'm sure ArmA 3 was at is apex (^^) if we look at the sale, when Altis Life became popular worldwide, so from this side I think the investors would have encouraged the dev to ""help"" the community!
Second, here we are not talking about a new feature for the game that really change the core etc... It's just some bugs probably linked to bad testing with different scenario / mods before the release. If the game would have gained some big changes, I'm sure the community would have take it much differently. If the QA would have tested it in a large variety of scenario / mods this wouldn't have happened in the first place...
rapid fable
I've not noticed single issue with this update in Vanilla Arma.
The sounds are a welcome improvement
bright pond
read this https://feedback.bistudio.com/T121736, it's vanilla ArmA (you are talking about only the campaign / included scenarios maybe?), and for me it's the responsibility of QA testing team to test all possible situation players will encounter playing the game, with or without mod.
rapid fable
Do you drive by any chance?
bright pond
I do not, why? x)
rapid fable
Cars are great examples. For example, Ford does not consult with Aftermarket parts companies before they make a design change. If you put off-road tires on your Porsche 911, it will not perform the same as the designers intended.
bright pond
Got it. So for you KoTH / Wasteland / Altis Life is not vanilla arma since it's not created by the dev' even tho it's not a mod, only a scenario. (actually depends on the server (some require mods) but you get what i'm saying)
Then vanilla ArmA is really limited...
nocturne obsidian
Let's face it
Without any mods arma would be long dead in the steam 2.99 sale bin
rapid fable
Vanilla == what came in the box when you bought it (or was d/l from wherever).
@nocturne obsidian I take it you've not been with the series for the last 15 years like I have then?
bright pond
Ok I understand, I was taking the minecraft (absolutely not ashamed :D) definition which is everything that run without a mod (including very customised / scripted server) is vanilla
nocturne obsidian
I've played arma for a very long time
And what you need to understand
Times change
rapid fable
Mods are great, but they don't "make the game"
nocturne obsidian
The entire premise changes so fast
People get board so fast these days
More selections
rapid fable
And they drop out just as fast
But the core players stay with the game
No matter what's popular today or tomorrow
nocturne obsidian
If mod makers never did anything and arma was vanila it would have less players than no man's sky
rapid fable
That's a very generalised statement. I think you'll find that it had quite a large ownership base before the game even shipped
bright pond
You think the core player of ArmA (or any other game) represent how many percent of the sale ? It's not with that, that a company will extend, take big AAA game, they base their business model on game that are abandoned 1 year later and it's them that make the more money (not saying it's great, just saying money fact)
rapid fable
First, I must apologise. I hadn't realised we're still in perf_prof. We can move this discussion to #general_chat_arma
bright pond
indeed
carmine stump
1.66, perf branch is still bad for me, I get 1/2 the frames I get with standard.
shell sierra
fix the damn footsteps though .....
empty goblet
1.66.139566 new PROFILING branch with PERFORMANCE binaries, v1, server and client, windows/linux
- fix some of the issues with eventhandlers
https://www.dropbox.com/sh/582opsto4mmr8d8/3BSy9PdRGm
https://drive.google.com/folderview?id=0B03-H4YIbhkFMUt5RzNqZjFlNGs
available via STEAMklient/STEAMcmd as branch too, read https://community.bistudio.com/wiki/Arma_3_Steam_Branches#Arma_3_Server
strong token
this build fixes
https://feedback.bistudio.com/T121730
tested and approved.
bright pond
Does it have to be client and server or only serverside works ?
uncut lily
been out of the loop for a while. did bohemia get any closer to finding the source to the crashy thingie ?
or the 3 fps bug
empty goblet
@uncut lily still being worked on, will get new debug traps for the memleak in profiling branch soon
low narwhal
Good news so far with 1.66, not had a 3fps bug however after 1 hour and 30 minutes my PC rebooted while playing arma.
shell mountain
running 1.66 profiling v1 on debian 8.5 x86/64, still getting fast fps slope down to 3-7fps
fallen quartz
@empty goblet using the profiling branch with the latest update lets you not launch the game with battleye activated. Windows eventviewer shows an error with event id 10016. Launching the game without battleye works. Only solution for me is to deactivate profiling branch right now.
empty goblet
1.66.139614 new PROFILING branch with PERFORMANCE binaries, v2, server and client, windows/linux
- sync with 1.66 hotfix and some crashfixes
https://www.dropbox.com/sh/582opsto4mmr8d8/3BSy9PdRGm
https://drive.google.com/folderview?id=0B03-H4YIbhkFMUt5RzNqZjFlNGs
available via STEAMklient/STEAMcmd as branch too, read https://community.bistudio.com/wiki/Arma_3_Steam_Branches#Arma_3_Server
empty goblet
1.66.139630 new PROFILING branch with PERFORMANCE binaries, v3, server and client, windows/linux
- more crashfixes
- fixed stuck VON transmit in lobby
- optimizations to reduce memory waste/leaks (scripting, functions etc.)
https://www.dropbox.com/sh/582opsto4mmr8d8/3BSy9PdRGm
https://drive.google.com/folderview?id=0B03-H4YIbhkFMUt5RzNqZjFlNGs
available via STEAMklient/STEAMcmd as branch too, read https://community.bistudio.com/wiki/Arma_3_Steam_Branches#Arma_3_Server
also i forgot to tell you, the increase to 3GB limit is back, hence -maxmem=3071 works in profiling and even stable
it's part of the new memory rework (related to 64bit cleanup of code/changes) so we re-enabled it
thus some feedback on that would be quite welcome too ...
(note: 3GB limit for maxmem is available only on 64bit OS, on 32bit OS it is still 2GB)
queen owl
The launcher still limits it to 2047. And I guess setting it manually in Steam doesn't bypass it? Unless with -nolauncher maybe?
Never mind, unchecking it in the launcher and using the Steam option works.
empty goblet
strange, my arma 3 launcher allows 3071, try pass the -maxmem=3071 from commandline / steam startup params
queen owl
yeah, that did it. But stable launcher definitely reverts to 2047 if I try anything higher.
empty goblet
ye, i'm sure @nocturne obsidian will sort it for next iteration ...
queen owl
A wild Wizard appears.
nocturne obsidian
It's preped and ready.
But someone forgot to mention it will be released with 1.66 (faceplam)
I expected the change to go live together with release of 64-bit game to dev-branch
nocturne obsidian
That's not for me to say ๐
empty goblet
not this week ๐
queen owl
So monday it is then. Got it.
Wait. Could -maxmem=3071 possibly help people with the "3fps" bug? Since it seems to happen when the RAM usage is near the old max of around 3gb?
nocturne obsidian
Changed to Launcher commited. That one will be in dev-branch on Monday for sure.
queen owl
๐
fathom cypress
Is that the 3FPS Bug caused by the sound?
Does the -maxmem=3071 help servers running on 64 bit OS ?
empty goblet
@fathom cypress it's not because it happens on server too ๐ it's just random speculation some people have w/o actual proof
@queen owl @fathom cypress the 3GB limit actually mean lead to more often crashes if the app/os exhaust memory faster (but the more reason we track it and fix it, so i want crashdumps)
just remember defining the maxmem is not defining the whole memory block engine uses, you change more of balance how engine does it with the ranges
viral portal
Seems to last longer before crash - dont get 6fps but black screen and only option to reboot pc
logs:
calm ember
The last time my game froze on a black screen, I could see, the process used VM =~3520ะะฑ. To 1.66, it has happened, often at 3300-3400ะะฑ.
uncut lily
@empty goblet nice, with the debug traps. btw. i've been running the 'system' malloc for over a month and getting exactly zero 3-fps bugs (although some dxgi crashes). tried the new tbb4 one the other day.. 3 fps bug within minutes
i don't know if it's the allocator or if it just exacerbates the issue having a faster one
and i regularly play for ~4 hours
per session
nocturne obsidian
Some Player on my Server say too me, if you have the FPS Bug -> set a Waypoint (Shift left + Left Mouse Button on Map) and the FPS Bug is gone.
Maybe someone can test it, because I do not have this bug myself.
mortal heath
Lol, never had any "3 fps" bugs
viral portal
good for you but that doesnt mean it isnt there.
mortal heath
What does it do?
sullen mist
@mortal heath You have just like 3 FPS in the game and you need to restart it
mortal heath
oh ok MR @sullen mist
lean dirge
Does alt tabbing have any effect on the FPS issue? I think a memory leak is causing it
@nocturne obsidian and others, have you tried flushing during gameplay? Shift + (num) "-" F LU SH
I think
viral portal
alt + tab have no effect - flush have no effect either - the memory allowed for the arma3.exe is used up, the only thing to do is restart arma
The latest perf3 have postponed the time till the issue arrives, but it will come.
Eagerly awaiting 64-bit guess that will fix the issue.
cursive vigil
64-bit on it's own may not fix the issue, it will just delay it further because there's more memory to leak before it runs out
atomic holly
64 bit will still help the game a LOT
cursive vigil
well Dwarden has repeatedly said people shouldn't get their hopes up and since he's testing the 64bit version ...
viral portal
64-bit might delay the 6-fps issue so long that no one will notice it any more. For me the perf3 delayed the prob so much that i actually will play the game again - before perf3 i could plat for aprox 45min-1hour - with perf3 i can play for maybe 2 hours.
empty goblet
1.66.139634 new PROFILING branch with PERFORMANCE binaries, v4, server and client, windows/linux
- more optimizations to reduce memory waste/leaks (scripting, functions etc.)
https://www.dropbox.com/sh/582opsto4mmr8d8/3BSy9PdRGm
https://drive.google.com/folderview?id=0B03-H4YIbhkFMUt5RzNqZjFlNGs
available via STEAMklient/STEAMcmd as branch too, read https://community.bistudio.com/wiki/Arma_3_Steam_Branches#Arma_3_Server
lean dirge
@viral portal haven't been able to test it but have you tried v4?
viral portal
im on it now - have had the first 6-fps - dont feel different from perf3 untill now
empty goblet
it's small steps and iterations toward decreasing the memory waste, will try get some merged soon
@viral portal using -maxmem=2047 or 3071 ?
shell mountain
@empty goblet v4 server on debian 8.5 amd64 with -maxmem=3071 loops between map init states, without maxmem runs fine
will have perf/fps feedback by tomorrow
viral portal
Tried both nothing and -maxmem=3071, noticeable better from perf2 to 3 but about the same from perf 3 to 4
referring to the 6-fps ussue only ๐
*issue
shell mountain
tried both v4 perf and prof, no perfomance changes, still fps issues serverside
empty goblet
v5 will come with another nice lil optimization ๐
nocturne obsidian
Can we get some idea of the scenario the optimisations are expected to help so we can test the actual impact?
empty goblet
small various optimizations, scripts, objects etc., but sometime they used often or everywhere so in the end it counts
nocturne obsidian
I think the limits for a 32 bit application like ArmA 3 are now achieved (graphics update and soundupdate -> higher Ram usage).
Well, Games like Subnautica (64 bit // Engine: Unity) use about 6 GB Ram, and have much less to offer than ArmA. => Is only a comparison (other engine different code...).
I think, the only way to reduce effective the RAM usage, is to delete the loaded textures and sounds after a while.
The question is whether this was ever planned -> is it possible at all?
But this would require more computing power again - higher data query of the HDD / SSD. And not to forget the loading times.
It is just the simple opinion of a normal player.
I hope for the 64 bit version.
queen owl
I think, the only way to reduce effective the RAM usage, is to delete the loaded textures and sounds after a while.
I'm sure that's how it works already. 2-3GB would fill up pretty quickly in a game like Arma.
empty goblet
1.66.139665 new PROFILING branch with PERFORMANCE binaries, v5, server and client, windows/linux
- optimizations to reduce memory waste/leaks (scripting, functions, objects etc.)
https://www.dropbox.com/sh/582opsto4mmr8d8/3BSy9PdRGm
https://drive.google.com/folderview?id=0B03-H4YIbhkFMUt5RzNqZjFlNGs
available via STEAMklient/STEAMcmd as branch too, read https://community.bistudio.com/wiki/Arma_3_Steam_Branches#Arma_3_Server
azure crest
@empty goblet we need to use the file arma3battelye with the perf/profiling ?
empty goblet
@azure crest it's just sync with the today's update for MAIN branch stable one
azure crest
okay thanks ๐
empty goblet
btw. with v5 you shall really observe tiny yet noticeable decrease in memory usage (in terms of less wasteful)
vast shore
Here i'm this is me
vast shore
@empty goblet =) It seems like i can't set 8191 to the -maxmem in the launcher
could you please paste the howto setup 64bit pictures
empty goblet
@stiff rover maxmem is twice in the arma 3 launcher, it depends what platform you select first (32 or 64 bit)
32-bit supports 2047 (on 32bit OS) and 3071 (on 64bit OS) while 64-bit supports 8191 max.value
vast shore
Actually found the picture.. but when I try to put more than 2000MB of memory to the launch 64bit value it resets it to 0
nocturne obsidian
@vast shore Sorry, it's typo there, the limit is at 8095
empty goblet
^ ๐ i spawned @nocturne obsidian so expect it fixed in next dev update
vast shore
8095 works :)
the tool tip says 20xxMB :)
@empty goblet =D
Battle eye fails tough :/
@empty goblet Inch told me that he will put a test koth server online as soon as he gets he's stuff done
vast shore
@nocturne obsidian Now that i have your attension how do I know it's 64bit build that is running, i mean the main menu doesn't tell me that
empty goblet
ye but there is no 64-bit server yet afaik (unless you use client binary with -server), there was no update today to dev
vast shore
attention, even.
queen owl
You see the build in windows task manager at least.
vast shore
@empty goblet Ah ok.. well maybe he will not then :) OOh i'm still running 32bit
nocturne obsidian
As Greenfist says; or switch it to a window mode (Alt+Enter) and then look at the name of the exe in the title.
empty goblet
@vast shore well i use process hacker so i've this additional column which tells me what type of binaries are actually in memory ๐ but what wizard said will work too ๐
also you can check RPT file log history for the executable name
vast shore
Oh i always have nologs. on
queen owl
By the way, how does the 64bit limit the use of my 8GB RAM? Should I use something like -maxmem=6144 to prevent the game from exhausting all of it?
Or does the game and Windows handle it by themselves just fine?
nocturne obsidian
Let the game handle it.
queen owl
Good. Thanks.
nocturne obsidian
@vast shore Eh, do you have BE enabled?
vast shore
wizard BE is erroring badly
14.12.2016 21:16:34 - BattlEye Service installation failed.
nocturne obsidian
So Launcher fallbacks to run without BE, right?
vast shore
it asks do you want to start without be
nocturne obsidian
What version of Launcher do you have? It's in the Options menu.
vast shore
1.5.139660
nocturne obsidian
There was this bug, when Launcher always run a 32-bit version when BE failed.
It was fixed in 139680
If you go to BattlEye tab and disable it, Launcher will start the correct executable.
vast shore
Yes now it works :)
I see the question was hard coded for the arma3.exe
So how to get latest launcher?
thx guys for the help :)
nocturne obsidian
Yeah, the fallback re-runs the play command without BE, but it ignored the platform and run the default.
It seems the todays update was cancelled. So it will on the dev-branch tommorow.
vast shore
151.80.99.57:2802 inch server.. :( but can't get on no battle eye
@nocturne obsidian Any change for you to put the latest launcher, out for grabs? Maybe dwardens dropbox?
nocturne obsidian
Maybe you misunderstood me. It can't fix BattlEye.
vast shore
ah ok
nocturne obsidian
The only bug in Launcher is that IF BattlEye fails to start the game, Launcher always started the 32-bit game without BattlEye. ๐
vast shore
Ok my bad i did misunderstood you
nocturne obsidian
But try to install it severe times more. I don't know what exactly is wrong with BE, but I've seen someone to report that it took several tries to install it.
vast shore
Was too excited =)
hallow yew
unpacking the perfv5 server.exe for Windows, slight typo :P
arma3servert_performance_166_139665_v5
empty goblet
always someone unhappy @hallow yew ๐
also now i can claim it was just your imagination ๐คฃ
shell mountain
sorry guys what about 64bit server for linux? also is client -server really usable?
empty goblet
no linux server yet, and yes the -server shall work nearly identical as dedicated server binary, just with more ๐
shell mountain
๐
viral portal
So there wont be a 64-bit 1.66 perf.exe ?
empty goblet
no, it needs dev branch, too many changes
viral portal
Rgr thx info
empty goblet
crashes, anyone ? ๐ฌ
stiff rover
vast shore
wait what?
queen owl
There can be only one. Now fight!
uncut flame
@empty goblet yet to have a crash.. will try harder later in my day perhaps. ๐
azure crest
have you problems with perf client v5 and -connect arg ? I stay in apex screen during the load ... i need to press escape et log on my serv with arma menu cc @empty goblet
silver roost
@empty goblet, where do you want crash files from performance build? Had 5 Crash from memory related errors tonight, sadly only one rpt cause i ran no log the rest of the times.
empty goblet
@silver roost sure, we want them to at least look, yet if you get repeatedly Out of memory crashes, best is to switch then to profiling binary (not performance which are default ones)
just beware the profiling binaries runs slower (contain debug layer same as development branch) but also contain special debug traps to help with memory leaks analysis
that don't mean they come with magical 'always catch and always figure it out' traps ... otherwise the OOM troubles would be long dead already
silver roost
Yeah will try next time. Can se i ran with no logs all games sadly. So will try that next time.
empty goblet
but if you can provide OOM crash dumps (rpt,mdmp,bidmp or rpt,bidmp) on profiling binaries, it's welcome and it may help more than from performance ones
silver roost
Il do that
empty goblet
running with nologs isn't that big issue, the crash details have exception, you can in most cases tell OOM crash by no MDMP generated, just BIDMP
silver roost
Okay il keep that in mind
empty goblet
updated #Arma3 startup commands BIKI doc with actual usage at 1.66.> -maxmem= , to put an end to another Urban Myth https://community.bistudio.com/wiki/Arma_3_Startup_Parameters#Performance
eager pasture
๐ ๐
64-bit Windows + 64-bit game: (physical memory*4)/5
What's the reasoning behind this particular formula?
empty goblet
updated -maxVRAM= description too just for sure ๐
empty goblet
1.66.139754 new PROFILING branch with PERFORMANCE binaries, v6, server and client, windows/linux
- crash fixes
- new server.cfg setting inside Mission Cycle: randomOrder=<Boolean>;
https://www.dropbox.com/sh/582opsto4mmr8d8/3BSy9PdRGm
https://drive.google.com/folderview?id=0B03-H4YIbhkFMUt5RzNqZjFlNGs
available via STEAMklient/STEAMcmd as branch too, read https://community.bistudio.com/wiki/Arma_3_Steam_Branches#Arma_3_Server
details: https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-34performance-binary34-feedback/?do=findComment&comment=3128781
eager pasture
+ new server.cfg setting inside Mission Cycle: randomOrder=<Boolean>;
New param. ๐ฎ
empty goblet
read the BIFORUM post, it's explained in example
eager pasture
BIForum feedback: https://forums.bistudio.com/topic/160288
Sorry, there is a problem
The page you requested does not exist
Error code: 1S160/2
๐
empty goblet
new forum software is ๐คฆ
cursive vigil
but it's new ... that means better, right?
at least that's what my Newspeak dictionary tells me
spiral pond
gj @empty goblet on the new parameter
atomic holly
ayy cant wait to try the new performance client
azure crest
the arg -password does not support with perf 5 and 6 ?
empty goblet
@odd oak all should work fine ... just check if you don't have too long commandline (steam may cut something if you have 100s mods)
azure crest
hm... no i just have -enableHT, and this line work fine in v4 perf @empty goblet
empty goblet
there is no change related to that and it works fine for me
distant oxide
Not sure if perf builds causing this or the 1.66 but my fps got considerably low, lost 10-15fps in heav gun fights, crowded areas where there is a lot of explosion going on. maybe its the malloc, i'll try and get back here.
nocturne obsidian
@empty goblet i have same problem since V5 and V5, commandline -password is broken...
My commandline : -connect=164.132.203.110 -port=2302 -password=agora -mod=@nocturne obsidian -noLogs -nosplash -nopause
The players must be connect manually
With the V4 it works very well, but not since the V5 and V6
rapid fable
Just had a continuous and flawless 6 hour event on Stable branch. No 3 FPS bug.
viral portal
i already had the 3 FPS bug many times with perf6, but i also use many mods.
shell mountain
linux server side i still go very low on fps 3-7, with any branch while running altis ๐ฆ
i guess we have important impact with the mods and mapping, but i do not remember such low perf all the time
silver roost
@rapid fable Was that with plain vanilla arma? or with how many mods and what type of game?
rapid fable
Apex +approx 13 mods , custom mission with ZGM functionality.
lime coral
Whats the new linux server appID ?
233800 worked a few times like a week ago, but not anymore.
[3:43:21][Step 1/2] Waiting for license info...OK
[13:43:21][Step 1/2] ERROR! Failed to install app '233800' (Invalid platform)
[13:43:21][Step 1/2] Process exited with code 0
tardy lion
233800 is the appid of the arma 3 tools, the server has appid 233780
lime coral
@tardy lion Thx, that explains a lot! i don't know how and why i got the tools appID. ๐
@tardy lion But, i wanted the perf_prof_branch version! not the normal!
@empty goblet could be that im blind but: https://community.bistudio.com/wiki/Arma_3_Steam_Branches#Arma_3_Server lists Arma3 Server AppID as 233800. And perf branch ID isn't specified.
dwarden : [ ... ] "STEAMcmd as branch too, read .... " the above posted link. What appID ? What am i missing?
empty goblet
@lime coral fixed the biki page, no idea why the tools appid was there
tardy lion
@lime coral to get a different branch with steamcmd you have to use app_update with -beta <branchname> and if the branch is passworded you need additionally -betapassword <branchpassword> - i.e. for the profiling branch of the Arma 3 Server:
app_update 233780 -beta profiling -betapassword CautionSpecialProfilingAndTestingBranchArma3
@empty goblet @tardy lion Thank you both very much!
lime coral
something is seriously messed up on my setup right now.
empty goblet
1.66.139792 new PROFILING branch with PERFORMANCE binaries, v7, server and client, windows/linux
- fixed -enableHT startup commandline param, linux (major) windows (minor), bonus benefit to AMD or CPU with many core users
- added support for Large / Huge Pages into allocator interface, included in default TBB4
- new startup commandline -hugepages (both client/server)
details: https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3130011
viral portal
@empty goblet only 3 files in v7 on dropbox - is it still uploading?
Guess it is 4 files now - ill be patient ๐
empty goblet
@viral portal i ran out of free space ๐ as usually too many updates for our community ๐คฃ
empty goblet
something to experiment as it may lead to performance gains https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3130040
random isle
did u read his message ?
- added support for Large / Huge Pages into allocator interface, included in default TBB4
INCLUDED IN DEFAULT TBB4
azure crest
yes i've read, thanks ^^
random isle
np
shell mountain
@empty goblet awesome news! gotta test immediately on prod server muahauh ๐
for now started with classic -enableHT and also -hugepages, was just wondering if that affects linux server build too
empty goblet
note: some more tidbits how tweak now https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3130040
@shell mountain i can't warrant you hugepages on linux yet ๐ (as it's still using the old jemalloc by default so i didn't investigate)
shell mountain
yeah was guessing so ๐
at least didnt segfault with inserting the parameter
would be so nice a more freshy jemalloc, but yeah tnx for all the progress anyway!
enableHT is improved on linux too?
empty goblet
yes, it was borked before ๐
shell mountain
well at a first glance looks already more distributed on the different cores and less single-core-peak
empty goblet
as i wrote in the followup post, best is to test, no commandline switch, vs -enableHT, vs cpuCount=yourcore count, vs cpuCount=littlemorethanyouhave ๐
shell mountain
yup, but being able to see difference only when loaded, need to cycle the different options in prod, i'll test and report about the others asap.
any tip for monitoring affinity behaviour on linux? still i preferr to look at real load behaviour
empty goblet
FIY: -enableHT is off by default (as it was until now) and -hugePages isn't available on linux yet and jemalloc isn't using hugepages yet either (only tbb4)
shell mountain
cat /proc/6552/status | grep Threads
Threads: 21
lol?
on i7 running with "-exThreads=1 -enableHT"
empty goblet
i hate when i forgot edit something, ofc the build is 1.66.139792
@turbid veldt remember the thread count is not exact, e.g. physx spawns own ones (we don't control that)
it's more about the absolute numbers
should be like this default < enableHT < cpuCount with custom values higher than default (which means no enableHT)
shell mountain
heheh, yeah i just recognized it too after freaking out not finding my server listed as 754 lol
running 139792, ok will try to explore those differences, mille grazie!
solar root
How do I know arma 3 is using huge pages?
empty goblet
you can hope, if you have the admin rights properly, the correct binary, the correct allocator, in correct places, with correct commandline ๐
shell mountain
๐น ๐น ๐
solar root
Do I need to have a coniguous chunk of memory to make huge pages work?
empty goblet
ye, but you on mercy of your OS๐คฃ
just try it, if you see some performance increase then be happy, if not well ... nothing to lose
solar root
OK
shell mountain
i7-jessie:# cat /proc/ARMAPID/status | grep Threads
default: "Threads: 25"
-enableHT: "Threads: 25"
-cpuCount=8 "Threads: 28"
-cpuCount=10 "Threads: 30"
all after misson loaded (but no heavy load), dunno if interesting...
merry shuttle
calm ember
@empty goblet Many thanks. This works well in the game smoothly.
Merry Christmas. ๐
We will be glad to see you after January 2nd. ๐
cyan dragon
Will try tonight the new branch and post result. Though our main issue isn't the performance, its the memory leak crashes on the client sides ๐ฆ
lost wagon
@merry shuttle Which monitoring tool do you use?
merry shuttle
@lost wagon
This one
https://forums.bistudio.com/topic/177406-yet-another-arma-benchmark/
;)
empty goblet
@merry shuttle with TBB4 i hope, system and jemalloc has no hugepages support
merry shuttle
@empty goblet I only use TBB4
I had the best performance with one 1.61 version where I had 38 FPS (not visible in the screenshot)
empty goblet
just FYI: #dev_rc_branch received it's performance binaries too, adding -enableHT and -cpuCount= fixes from 166.139792 plus -HugePages into DEV 1.66.139743 for both 32/64-bit binaries
cursive vigil
๐
empty goblet
counts the seconds before @cursive vigil asks 'where is the linux/mac port of this' ๐
cursive vigil
๐ค
merry shuttle
@empty goblet Hopefully the new dev version won't crash so offen... :)
empty goblet
well it's same build just with those changes added
random isle
167perf ?
empty goblet
ye ๐
random isle
isnt that Dev pref ?
distant oxide
okay so here's mine: http://imgur.com/a/gfhdv
My Specs are :
Fx 8350 @4.6Ghz
8GB DDR3 1866mhz (xmp)
msi gtx 970 gaming
asus m5a97 r2.0
for me cpucount = 10 increased my fps by 2
-hugepages -enableht didn't do so well against xtbbmalloc v3, which seems to be somehow superior above anything i've ever tried including the new version, renamed CMA(Community memory allocator)
What do you think? and as you can see in the screenshot that i lost my 5 fps somewhere along the road to 1.66 ๐ฆ
empty goblet
after new year the memory interface and sources will be available so i'm sure the CMA will adapt if needed
remember the performance loss migth be also drivers ... if you keep updating them
are you using CMA with preallocation or w/o ?
distant oxide
22:37:00: Init
22:37:00: Config:
User is admin: Yes
UseLargePages: Yes
SeLockMemoryPrivilege: Yes
SeIncreaseWorkingSetPrivilege: Yes
ForceMaxWorkingSet: Yes
HoldMemory: No
PreAllocBytes: 0.000
LockPages: No
22:37:00: Ext0SetPrivileges: 0, Error: 0
22:37:00: MinimumWorkingSetSize: 4096.000 mb
22:37:00: Ext0SetWorkingSet: 0, Error: 0
22:37:00: Using large pages
22:37:00: Init done
guess not ?
PreAllocBytes = 0
w/o
thin wyvern
xtbbmalloc is still the best memory allocator for my system https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3131493
amber totem
Merry Christmas to everyone in the Performance Branch. Always love to see the Arma 3 community coming together to solve a problem, especially a big one like performance. Much love to Darden!
Dwarden *
oblique panther
xtbbmalloc user here too @thin wyvern . But it comes and goes, since it is kinda experimental. But props going out to Dwarden and his team for their hard work!
nocturne obsidian
thanks
empty goblet
working on new @nocturne obsidian @viral portal
empty goblet
1.66.139937 new PROFILING branch with PERFORMANCE binaries, v8, server and client, windows/linux
- crashfix with lobby menu (may sort some trackir/freetrack crashes)
- yet another memory leak squashed
details: https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3138781
atomic holly
๐
eager pasture
Sweet.
eager pasture
Add -hugePages into the launch options under properties in Steam.
eager pasture
36.3 MB update to the profiling / performance branch (1.66.139938). patiently awaits change log
eager pasture
...
old tiger
why there is no support for -hugePages in the launcher?
eager pasture
Because it has only recently been added.
old tiger
I have it added to -par=par.txt lets see if that works
hollow wind
@eager pasture Thanks workingg great
empty goblet
1.66.139938 new PROFILING branch with PERFORMANCE binaries, v9, server and client, windows/linux
- fixed missing RPT-log header
details: https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3139161
magic valley
Is the headless client password issue fixed in this version?
lean dirge
in this most recent profiling branch or at least since v8 my fps is tanking with PP on. Bloom, PPAA, any of these, except for FSAA tanks my fps
I see a direct correlation from having FSAA on and forms of PP on top. Immediate drop from 50 fps to 30 fps in each case, on multiplayer server
empty goblet
nothing was done related to PP so w/e is your problem it's more likely drivers ๐
lean dirge
too much driver fun ๐ฆ
silent robin
@empty goblet have "we" (you guys at BI) got any closer figuring out the 3-5fps drop issue?
nocturne obsidian
Speaking of drivers. I am one of those having that 3 fps bug, although its 5 fps for me. Im now running an outdated nvidia driver from february 16 and i did not have any fps bugs since. But i havent had the patience to test it out for more than 3 hours though.. and is it true that radeon owners dont have the fps bug at all? it would narrow down the problem a lot if we found out, i think.
hallow yew
on an AMD r290x I had it very very rarely, maybe once in a year, with a 1070 I get it 3-4 times a week, one thing you could try is in Nvidia control panel, go to PhysX and make sure the processor is assigned to gpu and not auto
silver roost
I get the 3 fps thingy, every time arma wants to use more the 2950mb of ram, then it just spams my page file till it reaches 15gb, then crash.
viral portal
Yep Arma is just a nogo for some hardware configurations - and have been for months.
empty goblet
v10 is out ... it should readd what was removed as part of v9 fix but this time not broken ๐
eager pasture
English, please.
opaque wigeon
I NOT SPEAK ENGLISH
eager pasture
k ๐
hollow wind
Can someone explain what hugepages actually do?
eager pasture
hollow wind
Thanks!
empty goblet
1.66.139968 new PROFILING branch with PERFORMANCE binaries, v10, server and client, windows/linux
- readded what was reverted in v9 w/o breaking anything this time
details: https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3139898
nocturne obsidian
just a little update regarding using old nvidia drivers. for me it didnt work. while using a driver from feb 16, i still experienced the 3fps bug. im back to the newest drivers
also running perormance v8 at that time. didnt test with v10 yet
twin condor
Will the x64 build become the default .exe for stable branch some point it time?
silent nest
Nah, they just do it, because they are bored ๐
silver roost
@twin condor yeah sure. but no matter how much at the time, with the memory leak, it will crash at some point.
empty goblet
well, there is indication that some type of leak is caused by weird/incorrect use of script code ... more details later when it's confirmed
aka that don't mean it's also not revealing some other engine bug too ... but it seems it can be 'fixed' by less messy script code ...
twin condor
aw.. memory leaks are the worst to debug if they are in a script language inside a big project.. you need to write a script profiler for that...
does the script engine in arma use some sort of garbage collection? / GCRoots concept?
orchid garden
Which missions are most affected?
timid coyote
@empty goblet I didn't follow the whole cpuCount/enableHT/maxMem discussion in the thread. What was the result of it? No maxMem/maxVram anymore, and enableHT if you have a ht capable cpu (like i7)?
and what's with -hugepages?
and should -exThreads=1 still be used on servers?
lean dirge
Reporting back I overclocked the hell out of my CPU and performance is looking greater with profiling branch than stable
empty goblet
@lean dirge that's because it has more goodies merged ๐
lean dirge
๐๐๐ hype! I try to test all the different graphics settings around and I've just been much happier with a faster CPU from 2011
solar root
Where can I type addtional parameters in the launcher, such as -hugpages?
eternal thicket
You could always right click A3 on steam > properties > set launch options and add it there. If you have anything defined in those launch options you should also get a new "command line" field when you open Arma 3 Launcher
solar root
I used to be able to see the "command line" field. But now I can't.
atomic holly
hug those pages! @solar root
silver roost
Am i the only one where arma inceases my system page fil to an enourmous amout( over 15GB) And if i disable it, arma crashes after only an hours play
Only started noticing it after Perfomace v6, but cant be sure if it happens in stable. just seems like a huge amount
empty goblet
v12 is available btw.
nocturne obsidian
Okay guys, so how can I optimise my Arma 3 in the most efficient way possible? I'm on dev branch, i'm on some kind of performance binaries and I have (or so I think) all the correct things chosen.
My specs are 5960X @ 4.5GHz
32GB DDR4
GTX Titan X (Maxwell) Heavily overclocked
Only went from 37.4fps in 32bit YAAB to 41fps in 64bit YAAB
empty goblet
1.66.139995 new PROFILING branch with PERFORMANCE binaries, v12, server and client, windows/linux
- some more ogging in v11 (1.66.139981)
- fixed one of possible memory leaks and crashes
details: https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3141198
nocturne obsidian
@empty goblet Could you briefly explain what performance binaries mean?
empty goblet
@slender wharf imagine profiling branch it as experimental updates with changes which usually do run faster and more reliable than main branch stable has ๐
profiling branch has by default performance binaries as the normal user don't need slowdowns by debug layer ... and ofcourse it's MP compatible with MAIN branch
finite grotto
@empty goblet Are you saying I can implement Perf Bin on my servers as well as client? Been running like 6 Arma 3 servers for 2 years and 7 Argo servers since launch and I haven't come across this. #feelsashame.
empty goblet
of course, yes you can do that ... but because those're experimental it's always with some risk ๐
finite grotto
do you have a link for me to read up on how to implement it?
empty goblet
what about read ^ the post i linked
finite grotto
oh but you did say it is compatible and also more stable ๐
okie dokie. mucho gracias
btw
since you are on
I hav enot been able to log into BI forum
there seems ot be an issue each time I log in
I have cleared cookies and what not, have you heard anything about it?
since I got you here.
nevermind seems to be fixed ๐ havent log in for months because of some bug
silver roost
If i run profiling branch, but the game dont crash, just hangs forever. it do not generate a dumb file, or a log of any kind as far i can see.. ran it twice today with hangs and 3 fps. (related to the memory leak issues).
eager pasture
Where is the changelog for 1.66.139981?
nocturne obsidian
So anyone seen any performance improvements with the performance branch?
eager pasture
No... because it is not on 64-bit.
nocturne obsidian
Thought it was now?
nvm has there been any performance improvements? For clients?
empty goblet
@eager pasture there is single line entry changelog for the v11, v12 is latest
and you wrong, there are performance improvements in this branch too
eager pasture
1.66.139891 != 1.66.139981
@empty goblet, @nocturne obsidian was asking about 64-bit on perf/profiling.
For whatever reason the comment about 64-bit was deleted.
empty goblet
as i explained 64-bit will never be on 1.66 ... not worth the amount of work with merges
empty goblet
ye
i think it's obvious, considering on the dropbox is 139981 ๐
plus it's irrelevant as latest is 139995
eager pasture
Sorry, I'm a Steam Client pleb.
silver roost
Still experiencing 3fps on v12. sadly. Even though it seems its better, as the game somehow now can recover so i atleast can quit the game without killing the program.
flat current
Is it important that all clients connect with the perf client if the server is running it?
(dedicated server)
whole cloud
No. Well if you want the clients to experience the improvements of the perf versions then they have to run the perf client. But it's not required just because the server is running it
flat current
Cool, thanks. We have some folks who aren't having any problems... so why potentially introduce one, is what I'm thinking.
torn wasp
do the perf and prof not working with battleeye?
empty goblet
@torn wasp should work fine
cyan dragon
Hey Dwarden, I got around like 40 people having performance problems and steam null invalid.. Recommend to run the perf branch client side also??
nocturne obsidian
Hello everybody our dedicated server is in 1.66.139586 it's the last update no ?
How to install as a client (last post ) https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?page=77#comment-3141198
pleae ?
nocturne obsidian
Ok thx if install this just as client and not on our server It corrects the problem of 3 FPS ?
I am French and it is complicated for me to inquire, thx for you're help
eager pasture
@nocturne obsidian, to update to perf/profiling branch for Arma 3 Server via SteamCMD use:
app_update 233780 -beta profiling -betapassword CautionSpecialProfilingAndTestingBranchArma3 validate
For the client enter CautionSpecialProfilingAndTestingBranchArma3 in the box under "Betas" โ "Arma 3 - Properties" in Steam client.
nocturne obsidian
More Thx !!If I do it on the client and not on the server it will work?
eager pasture
Yes.
nocturne obsidian
Then why update server ? ๐ I'm curious
empty goblet
because some fixes are aimed on both client/server some server some client @nocturne obsidian
overally, engine fix or optimization is usually all of them ๐
nocturne obsidian
ok thx then just client update and no more problems with 3fps bug ?
empty goblet
i never said that, i said 'another memory leak bug and crash issue is fixed'
we already fixed multiple ones ...
nocturne obsidian
Okay it was a question, more thx ! last question EtA before officiall 64 bit update ?
empty goblet
when ready, irrelevant to profiling branch / performance binaries, as i explained endlessly it will be in DEV branch as long as possible until we finetune it to be enough stable
and now something important
ok, here is script-code what caused known memory leak / overloaded netlayer and lead to server freeze / stall later
[] spawn {
while {true} do {
if (!isNull objectParent player) then {
detach player;
}; }; };
and this was workaround script-code attempt to address it (no more leak/perf issues nor crash)
[] spawn {
while {true} do {
if (!isNull objectParent player) then {
if (!isNull attachedTo player) then {detach player};
}; }; };
since new Profiling branch 1.66.139995 (and it's performance binaries) in #perf_prof_branch
this is covered and FIXED at low-level within engine itself, to prevent leak, stop performance/netcode issues etc.
yet,
please take look into your script code for similar cases of 'weird' script-code related to different commands as issues may still exist
nocturne obsidian
Sorry, but I do not understand, but thank you for trying to explain
empty goblet
similar bug may exist for different command, hence i share it so creative community can take look and help narrow it down ...
silver roost
@empty goblet I have ran profiling a few times now, but when the game do not crash, but still produce the 3fps error. How can i help, for cant seem to find any logs then
empty goblet
well i was nearly sure it's not the leak we looking for ... it was another one ๐
nocturne obsidian
Thank you for all of your continuous hard work on Arma and especially working on the memory leak, known as the 3-5fps bug!
cyan dragon
fyi I have got the 3-5 fps bug a few times, once just loading in. Another time when texture not refreshing enough it then got stuck and 3-5 fps kicked in
magic valley
Quick observational report ref: fps bug.
Playing on our dedicated server last night. Server is running v12 performance binary as are all the connected clients.
Playing in window mode on client.
I tabbed out the game after about 2 and a half hours of really smooth fps gameplay never dropping below 35fps on client(apart from when I initially used the rangefinder where it bombed to 15fps then recovered).
I tabbed back in after about another 2 hours only to find that my client had the fps bug!.
cyan dragon
Jman, long time no seen. Anything extraordinarie in the RPT?
magic valley
I'm always lurking ๐ I'm afraid not no.
cyan dragon
What causes it is the question ๐ฆ
hallow yew
it seems to be related to ADS/rangefinders/binos
empty goblet
@magic valley make sure the server runs v12
magic valley
Yes it is sorry I meant to say 12
empty goblet
which means the hunt then continues ... that wasn't the leak we were looking for ...
magic valley
Indeed
nocturne obsidian
Where does one download these binaries?
eager pasture
Click on the pin in the top right to navigate to the latest version.
nocturne obsidian
Ohhh "pinned messages"
I see.
Wonder what that was for xD
@eager pasture Perf 1.67 ?
eager pasture
The versions should be even numbers.
eager pasture
๐ค
@empty goblet, can explain; I cannot.
I'm using 1.66.139995 (v12) which is the latest available via Steam client.
nocturne obsidian
Steam client?
queen owl
open the game's properties in Steam library. Insert the code 'CautionSpecialProfilingAndTestingBranchArma3' on the BETAS tab.
Steam will then update to the latest perf binary whenever one comes out. Which is usually very soon after Dwarden puts a new one in his dropbox/google drive
eager pasture
@nocturne obsidian http://store.steampowered.com/about/ and as @queen owl said.
eager pasture
Performance and profiling for client and servers (1.66 v12) - https://drive.google.com/drive/folders/0B03-H4YIbhkFbmsyVTNfVm9WdTQ
woven loom
So, to confirm, I can run these binaries with stable branch servers, correct?
eager pasture
Mix-matching between client and server versions of main/stable and perf/prof is okay as I understand it.
eager pasture
Other than the usual arma3.rpt, arma3.bidmp and arma3.mdmp?
eager pasture
Little more info here - https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/
PERF = STABLE branch compatible, no debug layer, no #captureFrame nor diag_captureFrame
PROF = STABLE branch compatible, with debug layer, with #captureFrame / diag_captureFrame
I see, probably want profiling branch then
lean dirge
Never had fps bug, but I am best PC man of all time too
hearty sierra
Well, if the performance binary was supposed to fix the out of memory error: It did not
(a.k.a low FPS bug)
thin wyvern
@hearty sierra did you try -hugePages startup parameter https://community.bistudio.com/wiki/Arma_3_Startup_Parameters
hollow wind
Perf did help the 3fps bug a bit
rather than crashing right a way
it gives us a little time to bail
anway what does PC stand for?
flat current
Are both the performance and profiling exes referred to as "profiling branch"? Can I use -hugepages with performance as well, or just specifically with profiling?
eager pasture
Yes and both binaries.
flat current
Thanks. ๐
rapid kettle
@empty goblet did some testing using various values for -cpuCount, -exThreads, -hugepages and -enableHT on profiling branch, results and system specs can be found here: https://docs.google.com/spreadsheets/d/1k82m6mJRR5iCGX9YGll7eogO-678TCRyafv7JWD4JJQ/edit?usp=sharing
long hemlock
a whole lot of work for you lol
rapid kettle
yeah that was a few hours of benchmarking, but at least it shows there is possibly a minor benefit to using -cpuCount higher than your physical cores + logical cores
as well as other conclusions, but yeah still worth
long hemlock
did you OC your CPU and disable hyper threading?
rapid kettle
it was OC'd to 4.6GHz, and no, HT was on the whole time, I was mainly testing the hypothesis in this post: https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3130040
long hemlock
meh, I'll just wait for a stable release and restart Arma every 2 hours to avoid the memory crash.
rapid kettle
yeah, still a decent workaround
hearty sierra
@thin wyvern Haven't try that yet, I'll try next time I play for a longer time
random isle
@empty goblet when is this memory bug going to get fixed, i get slideshow / crash every 20-40 min, been like this for what- 4 months already ?
silver roost
Temporary fixed my 3 fps bug. I reformatted the pc. And havent had it in a week. Same mods same missions. It will probably start again at some point. But damn its nice to not crash every two hours
empty goblet
it could be some of those 'sure-tweak-how-improve-windows-performance' which was done years ago and one forgot ...
but that don't explain ofcourse why it happens to users on fresh W10/8.1
mumbles something about kingdom and half horse for details on trigger leading memory starvation and crash ...
cursive vigil
nor why it sometimes, though more rarely, happens for users on Linux
empty goblet
so, anyone getting crashes on 1.66.139995 yet ?
nocturne obsidian
thats what? v12? no crash but 3fps after 2 hours. at the exact same time another player got it, by the way.
viral portal
v12 like useual mem crashes
whole cloud
diag_logSlowFrame only captures once after the script call when the slowFrame is detected.. And there is no way to detect scriptside that it already captured a slowFrame... correct?
So if i wanna setup a script that logs all slow Frames in a full mission i have to call diag_logSlowFrame repeatedly every few seconds?
silver roost
Have had two crashes on v12 performance, after reformat still none on clean stable.
heady parcel
v12 doesnt crash for me, fps bug, game freezes wait it out eventually have to force close.
random isle
im on V12 and its behaving same, 5 fps after 20-40 min then if you wait it will crash.
hearty sierra
@empty goblet Yes, I got a crash with 1.66.139995 same as always, low fps followed by out of memory error
eager pasture
I'm sure he'd love a crash dump, @hearty sierra. ๐
hearty sierra
I'm sure, but I don't have one, and it's the exact same crash as always, so I would guess there are lots of crash dumps already
empty goblet
well if you have crashes then all files (mdmp,bidmp,rpt) still welcome (7zip it, send me pm with url download)
also if you have time try it run on profiling binary as that has additional logging and traps
hearty sierra
We're playing tonight, I'll try the profiling binary
cyan dragon
lol the FPS bug is so freaking weird for example
I can instantly get it during heavy firefight
when im being supressed.
Then I can get it when I suddenly join the server
hollow wind
This bug never happens when there are less ppl
and no Explosion or Heavy Fire fights
maybe there is something wrong with explosion script
effects or something
cursive vigil
right, explosions or very heavy gunfire have always figured into it for most people. This is why one theory was that it was related to the audio overhaul since the bug first appeared around the same time
though Dwarden has dismissed that possibility
long hemlock
I always get the 3 fps when fast-travelling the map. i.e. two fly-overs in the F-22 etc
eager pasture
Game crash (profiling) - https://feedback.bistudio.com/T123006
viral portal
@long hemlock yes fast traveling on map makes avail ram to be used fast.-I dont even bother fly any more, a full chopper with players + 3 fps crash will prob give you enough teamkills to be banned from server.
hearty sierra
@empty goblet Re FPS bug... one of our guys says that if it happens for him, he ALT-Tabs back to desktop and then back to the game, That solves it for him (fullscreen only, not windowed)
Looks like it's leaking textures/video ram
viral portal
alt+tab do not help here - for me it might bring game back to life, but only for a few mins, and second time i get it, i wont be able to alt+tab again.
hearty sierra
@viral portal Yeah, but the point is, this points to a video ram issue as the cause of the low fps/out of memory problem. Of course it has to be fixed, but it's just an observation that alt+tab brings it back (even for a bit)
viral portal
I do belive its a problem with normal ram rather than video ram
long hemlock
alt-tabbing will crash my game entirely. I just ESC and then alt + F4 out of the game. that way I don't need to reset my whole system
viral portal
Gotta try the alt f4
amber totem
Definitely normal ram
It's been an issue plaguing me and my server for eons
I pray every night to the glorious Dwarden fixes will come
cobalt leaf
Oh, a problem that I have!
Yeah, I have everything above. I usually have about 5 or 6 gigs of ram when I start arma, but by the time I have the major FPS drops and game crashes (Unable to init Dxi, etc) it's around 8.9 gigs
This is on the tbb4malloc
Also, possibly unrelated, I noticed better performance disabling exthreads (But I have AMD, which might be why)
I'm guessing memory leak
Shift+- FLUSH doesnt even help too much.
random isle
yep alt tab kills my game 100% of time, i have to run it as a window ๐
silver roost
^ gotta love how arma is kinda the "special" kid in the class
eager pasture
@random isle, do you use the -noPause launch parameter? https://community.bistudio.com/wiki/Arma_3_Startup_Parameters#Developer_Options
random isle
@stoic ivy hmm ill try thanks
viral portal
Re 6fps bug - i def get best result (can play the longest time before 6-fps) by using -malloc=system (win10 64bit)
empty goblet
1.66.140166 new PROFILING branch with PERFORMANCE binaries, v13, server and client, windows/linux
- fix for issues on Linux server with too many PBO (unable to connect to afterward)
- multiple crash fixes
- fixed sounds vanishing in SP while opening map
- EnableEnvironment improvement (possible to disable ambient life and environmental sounds separately)
- more tweaks related to previous fixes of detach/attachTo
details: https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3145555
wanton lynx
Just to echo what I said in #server_linux I personally still haven't had any luck with the linux PBO limit
empty goblet
@wanton lynx test the new version ๐
wanton lynx
uh, yea, I did
boreal wigeon
@empty goblet what ever you did with the pbo limit thing it now allows me to load less pbos? server doesn't start up with version 13, v12 worked fine.
empty goblet
@boreal wigeon that's strangely worrysome
same as with theowningone i will need exact details about the nix server
heady parcel
droped in v13 on centos7 couldnt connect to the server. Switched back to v12 could connect : Linux chicago 3.10.0-514.6.1.el7.x86_64 #1 SMP Wed Jan 18 13:06:36 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux
empty goblet
๐ฆ ๐ข
whole cloud
What are the pbo counts of the people having problems?
If you report anything about PBO limit please tell OS version (uname -a should do it) and count of PBOs including or not including Arma vanilla pbos
wanton lynx
So, I haven't actually tested my box with the previous workable number on v13 yet, going to test that now
I can also confirm at this time that the PBO amount that previously worked for me does not in v13
I'm running Ubuntu 12.04, although it hasn't been updated recently
Full uname -a output is:
Linux dicgaming 3.8.0-29-generic #42~precise1-Ubuntu SMP Wed Aug 14 16:19:23 UTC 2013 x86_64 x86_64 x86_64 GNU/Linux
For testing PBO count I've been using this one-liner:
lsof -a -p `pgrep 'arma3server[^\.]*$'`| grep '\.[e|p]bo' | grep -v Dropbox | wc -l
(We symlink our mpmissions to dropbox, grep -v Dropbox removes those entries from that list)
That metric there return for me, when it's working, at or below 944
if it hits 945, the server becomes unreachable
some ulimit info: http://pastebin.com/fPkM3EyX
let me know if you can think of anything else helpful
empty goblet
ok, ty, will know more thru workweek
empty goblet
anyone had v13 (140166) crash(es) over weekend ?
viral portal
For me nothing changed in v13 in regards to 6-fps/memory crashes
empty goblet
no need to post logs here ... (nothing useful in that text block either)
spiral pond
Tweaked: Terrains in the Main Editor Menu are now sorted alphabetically
๐ฎ
fallen oar
azure crest
hi, i use the 13v perf i have cpuCount=10 and i look this : http://i.imgur.com/FXn4zFb.png
empty goblet
@wanton lynx @heady parcel v14 linux ready (1.66.140235) , should fix the pbo limit issue finally
empty goblet
1.66.140235 new PROFILING branch with PERFORMANCE binaries, v14, server and client, windows/linux
- new fix for issues on Linux server with too many PBO (unable to connect to afterward)
- tweaks for maxPing, MaxDesync, MaxPacketloss (better median etc.)
details: https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3148283
wanton lynx
Checking now
I can confirm at this time that it appears to be working on my end
I want to do more testing a game later this week to be sure, but I'm feel fairly confident in calling this issue resolved
Thanks so much @empty goblet (and any BI devs lurking out there)
whole cloud
Can you share what caused that problem? :x I'm interested
boreal wigeon
@empty goblet running v14 now and it works, thx for the fix.
ill test running more pbos now.
empty goblet
remember to raise your ulimit if you can ...
heady parcel
I was able to connect, testing more pbos in a few
hollow wind
Havent experienced the 3fps bug again for a while
But some of the members are still experiencing it while watching artillery strikes
A10 or just huge explosions
Still Great work
Getting better xD
empty goblet
just to be sure , anyone here who had no luck with the new linux fix ? (aka too many pbo still trouble)
boreal wigeon
@empty goblet running over 1000 pbos with the filler_mod and it seems to work, ill try something else right now where I had issues with in the past.
Yea, it seems like you guys fixed it, ran an old scenario where it didn't run, and just tried it and it worked.
empty goblet
1.66.140270 new PROFILING branch with PERFORMANCE binaries, v15, server and client, windows/linux
- crash fix related to items manipulation
details:
https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3149407
note: this was abused ITW to crash servers, so i suggest you adopt this build on your hosting setups
empty goblet
1.66.140303 new PROFILING branch with PERFORMANCE binaries, v16, server and client, windows/linux
- another server crash fixed
details:
https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3150306
nocturne obsidian
How I can reproduce the 3 FPS bug: Starting A3(v15)32bit. Loading Tanoa with zeus. View distance and obj view distance at max. Put down lots and lots of Fuel Trucks. Blow them up with rock or something while watching in zeus. Repeat for 5 Minutes. Do not remove wrecks. Then move camera around tanoa. at least at this point i will receive the 3fps bug. Trees and buildings disappear and just the ones near the fuel trucks stay visible. sometimes leads to a crash but most of the time the game keeps running with 3fps until i close the game via menu or alt f4.
i uploaded some logfiles: https://www.dropbox.com/sh/pbgmil5e282p4d7/AAADa7phGbKKbxf5cDiKbrO3a?dl=0
included are the a3 .rpt and .bidmp but also dxdiag and procdump before 3fpsbug
couldnt get the procdump during 3fpsbug
hope this helps somehow
empty goblet
what about make repro mission ๐ (you can preplace the trucks, you can detonate them e.g. via trigger or auto scripted and you can prescript the camera flight path too)
nocturne obsidian
im sorry, but i am not capable of doing so since i never got a hand of the editor to do such things. also i was using the mods and the servers of my community.
empty goblet
because 'flying over tanoa' is quite too generic for repro
nocturne obsidian
while i tried to reproduce it, it didnt care in which direction i moved the camera. after 50 or 100 m the 3fps issue became presend
im sorry for not having the skills to do such a repro mission. but maybe some other dude who is reading here is going to do it in my place? maybe with vanilla a3, too? ๐
uncut lily
@empty goblet it's that random. you can't do an exact repro. basically you need to force the game to load a lot of terrain from altis or tanoa at once. i don't know about the innards of how it uses the memory since i don't know how to peep into the stack/heap, but when you do that and at the same time monitor ram usage on the arma process you will be able to see that "private bytes" increases. it never decreases
so the game doesn't return memory to windows. maybe it's designed that way, but then i just think it's odd it doesn't allocate ~3.5GB from the very beginning
an easy repro for us that have the 3 fps issue a lot is just getting in a fast jet, upping the view distance and flying around. guaranteed 3 fps + crash within 30 minutes
as a sidenote. i can play on 110 player KOTH servers without issues if i just set the view distance to 1000m
then it never crashes and it never goes into 3 fps
also, since KOTH allows you to adjust view distance easier than the ingame menus you can up it to 4km. at that view distance i will get a 3 fps mode. at 2km, it will take longer to get the 3 fps. another thing you can do with that is increase it to 4km view distance, then reducing it to 1km again. it will still crash because arma will not unload the terrain
so i'm guessing the issue is either terrain, building models, vegetation or decals not deleting itself from memory
i.e garbage collection not working
i'd speculate that some classes are being 'touched' so the garbage collector doesn't see them as old and deletable
standard programming error for many of software products i've worked with ๐
i assume you do garbage collection on classes since it's a networked game service right?
actually, it would be nice if i could see how the memory deletion occurs in logs. is that possible?
empty goblet
@uncut lily i think i've got some partial-repro and especially tracking some event which happens always but i don't want to claim 'we see the bug' yet
uncut lily
@empty goblet if it was easy, it wouldn'tbe fun would it ๐
boreal wigeon
v17? whats new?
empty goblet
1.66.140341 new PROFILING branch with PERFORMANCE binaries, v17, server and client, windows/linux
- fixed Crash when equipping Uniforms, Vests, Headgear or Backpacks for a different skeleton
- fixed zoom sometimes not working properly in multi-player
details:
https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3151359
loud cradle
steam auth server problems - blame steam
nocturne obsidian
check #announcements jasper
boreal wigeon
Yea saw that was just hoping on a update ๐
whole cloud
or #arma3_troubleshooting most chatter about that seems to be there
grim wren
@empty goblet Any news about the 3FPS bug?
Anything new we can do to help?
I'd rather hear smth bad than nothing, this sh*t is killing arma for me right now.
cursive vigil
@grim wren https://feedback.bistudio.com/T123238
empty goblet
just several indicies and possible trails ... nothing worth to mention yet
whole cloud
@empty goblet Can I get something back for finding that Eventhandler thingy? For example the script debugger being enabled in prof builds? wink wink Maybe I can find more such problems using that ๐
empty goblet
to be fair, you expect something to exists, what's not even there ...
i guess you mismatching some features existing only in builds used by the other company, doing serious simulations
whole cloud
So the scriptDebugger thingy was completly removed? :/ No it wasn't! I could still find the BIDebugEngine interface in the latest binary. But I guess thats a no :x
empty goblet
i would say there never was one in first place ๐
whole cloud
๐ But why. I think it's very useful. And people are having a hard time writing their own script Emulator and debugger when there theoretically was one they could theoretically maybe just use
spiral pond
seems like Ondrej or other programmers kept some secrets from Dwarden after all ๐
empty goblet
@spiral pond again you assume wrongly ... for now there is nothing ๐
silent nest
" for now " - HL3 confirmed
flat current
Heheh... HL3.
thin wyvern
@empty goblet When and why -bandwidthAlg=2 param becoming obsolete?
hallow yew
@thin wyvern I think that was used as a test alternative networking method for Arma 2, I tested it for a while back then, i'm not sure if it has any effect in Arma 3, there's not any reports of it's effectiveness and Bohemia probably don't have more intel than this
it's just different and experimental algorithm
it defines way how game calculates priorities of queued updates in traffic channels```We can use that commandline for A3 too because we have that param in A3 wiki startup parameters but they remove it.
And arma3.exe still have it in.
server config= bandwidthAlg= pid= client port= ip= password= autoinit bepath= xa2sound init= showscripterrors netlog physics= noland world= enableHT cpucount= nosound d3dNoLock d3dDynCB= d3dNoMultiCB d3dCreateSingleThreaded d3dCreateNoT
whole cloud
Can confirm bandwidthAlg is still working and doing stuff in 1.66. Maybe it's removed in 1.68 and that's why it was removed from the wiki
cursive vigil
@thin wyvern But did it actually do anything? What were your reasons for using it?
empty goblet
@hybrid sparrow when i finally manage to get it adopted as default ๐
btw. -bandwidthAlg=2 is still recommended cause it's way better than default for any server with more players
thin wyvern
@empty goblet Can you then please put back that parameter info in Arma 2/3 wiki startup parameter page ๐
empty goblet
it never was on that page in first place ...
thin wyvern
On which page then I can't remember ๐
nocturne obsidian
hello I have a little problem ;/ I downloaded the Arma III performance client, but there is an error "PxFoundation wasn't created" anyone have solution for this?
empty goblet
1.66.140454 new PROFILING branch with PERFORMANCE binaries, v18, server and client, windows/linux
- fixed some Crashes
- fixed setUnitPos vs Careless AI soldiers on patrol
details:
https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3153706
random isle
@empty goblet Careless AI is that anything to do with remote ai go into hold fire mode?
woven loom
@random isle probably wrong channel, if it's dev branch related see #dev_rc_branch
empty goblet
so, anyone with some disastrous crashes on latest PROFILING branch PROFILING or PERFORMANCE binaries ? (other than the leak 3fps nightmare)
woven loom
@empty goblet what should I do if i experience low perf with profiling branch? As in, how do i capture data? This will be in MP so i won't have console access
whole cloud
@woven loom Get yourself some Mod to execute scripts ingame. You can also open up the Dev console to logged in Admins in MP. and then just execute diag_captureFrame 1
For more info https://community.bistudio.com/wiki/Performance_Profiling
woven loom
Ah, thanks. Although didn't really need it, profiling branch held perf quite well
Other people had to restart, I didn't
So they've probably fixed whatever is giving us issues at the moment
Will try again on tuesday with mods
woven loom
Also, i'm not admin, although I could try on the mission making server...
(single player usualyl doesn't give me issues)
silent robin
@empty goblet sent you a pm regarding the 3fps stuff.
woven loom
so there's still RPT spam in profiling branch, though as I noted before performance held
20:30:46 Attempt to override final function - rscunitinfo_script
20:35:12 Strange convex component149 in a3\plants_f\tree\t_pinuss2s_b_f.p3d:geometryView
20:35:12 Strange convex component65 in a3\rocks_f\sharp\sharprock_wallh.p3d:geometryFire
20:35:14 Strange convex component06 in a3\rocks_f\sharp\sharprock_monolith.p3d:geometryFire
20:35:14 Strange convex component18 in a3\rocks_f\sharp\sharprock_monolith.p3d:geometryFire```20:35:17 Dimensions in class I_Mortar_01_F should be an array of size 2.
20:54:04 Client: Object 2:3506 (type Type_121) not found.
tons of
20:54:05 Object WeaponHolderSimulated can't have zero or negative mass! a3\weapons_f\dummyweapon.p3d
20:57:48 - network id 2:4773
20:57:48 - person Mylonaki
20:57:48 - dead```21:01:24 Sound: Error: Multi-stream: Load failed, stream a3\sounds_f\characters\crawl\metalplate_crawl_4.wss not found !!!```21:27:04 Error: Number of retries exceeded for releasing physx actor. Released now.
21:56:21 WARNING: Function 'name' - B NATO A1 -:4 REMOTE (UnitNATO_A1_AT) has empty name
That's the most frequent kinds of messages that are repeated in RPT
woven loom
okay so i was trying out diag_captureSlowFrame on the community test server
I get this
Is this not how it's supposed to be used?
whole cloud
Are you sure you are running the profling executable? It should show dev build at the bottom right I think.
These commands will be unknown to the game when you are not running that executable. That's why your script errors.
whole cloud
Dunno. I don't use the Steam branch nor the Launcher ^^
eager pasture
The Steam profiling branch uses the performance binaries by default; not the profiling binaries. https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/
PROF/PERF client and server are available here :
https://www.dropbox.com/sh/582opsto4mmr8d8/3BSy9PdRGm
https://drive.google.com/folderview?id=0B03-H4YIbhkFMUt5RzNqZjFlNGs
remember to backup arma3.exe / arma3server.exe before replacing with PROF/PERF binaries
note that above URLs for manual download might be overloaded by demand, in such case
you can always use the STEAMclient to switch to profiling branch
(performance binaries are used by default so nothing to worry about)
@icy tulip Here.
icy tulip
cheers @silent nest, not for me although I'll I might try it out. it did fix the zoom just felt it effected rendering of objects
nocturne obsidian
will there be more perf updates or will the next update be the 64bit release?
empty goblet
will see, depends if i find anything worthwhile to merge ... of course there will be more perf after 1.68 main hits stable
nocturne obsidian
alright. looking forward to this. and thank you for your work ๐
empty goblet
q, anyone using the max* (ping,packetloss,desync) and disconnectTimeout settings in server.cfg ? https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?page=61#comment-3035423
i would like to hear some feedback (for 2 weeks there was improvement in the calculations so the erroreous values shall happen less often (not the 100k+ desync, that's different problem))
lost wagon
@empty goblet Currently not, due to the errors with kicking a lot of players if there is a short desync/server lag. But if I have some time, I will test that again!
empty goblet
@lost wagon do not use the maxdesync for kick, just logging as that's still unresolved for the max value i was more interested into feedback about maxping, maxpacketloss, disconnectTimeout
lost wagon
@empty goblet
My current Config:
maxPing = 400;
maxDesync = 90000;
maxPacketloss = 30;
disconnectTimeout = 10;
kickClientsOnSlowNetwork[] = { 0, 0, 0, 1 };
Yesterday (21.02.2017) from 18:00h to 04:00h the server logs just show one person with ping too high (674ms). Altis Wasteland 70 Player. Nothing else.
Seems to work smooth. Seems that also the high ping during mission download is fixed. I really consider to enable the kicks for slow network now (except the maxdesync).
If you need any logs, feel free to contact me!
empty goblet
dT on 10 may cause some kick issues - if it does then use 11 as in my test that worked always
(overally applies the same system try 5 (ultra fast network) and then slowly increase until it fits your userbase)
lost wagon
dT raised no problem for my servers. 10 works fine for me. I can try values lower than this, if you need the test results for dT lower than 10. Anyway, 10 seems to work smooth for me since a long time.
empty goblet
good to hear
daring sinew
It appears the perf branch has dumped my server config setting of default difficulty?
daring sinew
I'm also noticing some players with Integer.MIN_VALUE as their ping somehow...
empty goblet
@daring sinew should be fixed in DEV/maybe RC ... if you talk about the value in GUI
empty goblet
got crashes from v18 ? keep PM me those crashdumps ... if you did already thanks again ...
waxen burrow
@empty goblet sent you dumps gathered from 4 different clients. All running v18. Good luck!
empty goblet
i know @waxen burrow
woven loom
I actually did get a crash, when I tried to open a mission in the 3D editor with perf (happened one out of two times i tried, then i switched back to stable)
nocturne obsidian
v18
disconnectTimeout = 30;
Slow network for player name='x', ID=y, steamID=z. DisconnectTimeout too high (15/30000 ms)
Slow network for player name='x', ID=y, steamID=z. DisconnectTimeout too high (26/30000 ms)
Slow network for player name='x', ID=y, steamID=z. DisconnectTimeout too high (54/30000 ms)
Slow network for player name='x', ID=y, steamID=z. DisconnectTimeout too high (7/30000 ms)
Different players.
30 seconds - 30000 ms but what is the value 15 / 26 / 54 / 7 - ms?
@empty goblet
empty goblet
strange, probably bug ... will take look into it
empty goblet
@nocturne obsidian what were the other server settings related to client network connectivity state , if i may ask (the max* and log/kick one)
nocturne obsidian
maxPing = 256;
maxDesync = 80000;
maxPacketloss = 50;
disconnectTimeout = 30;
kickClientsOnSlowNetwork[] = { 1, 0, 0, 0 };
empty goblet
ty
empty goblet
1.66.140687 new PROFILING branch with PERFORMANCE binaries, v19, server and client, windows/linux
- fixed more Crashes
- fixed lineIntersectSurfaces vs primary LOD and rounding, extended with optional bool to allow return of same object intersection
- fixed dead gunners to be thrown out when swapping position into theirs
details:
https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3159505
nocturne obsidian
perf19 is empty, i guess there was anything wrong? ๐
lost wagon
Google drive and Dropbox are empty O_o Anyway steam perf branch were updated
empty goblet
ye, fixed, sorry @nocturne obsidian and @lost wagon
got sidetracked by other important stuff
queen owl
Stuff like a big stable update?
nocturne obsidian
np got it thru steam already ๐
random isle
@queen owl define stable ๐
empty goblet
more important stuff than stable stuff ๐
woven loom
#3fpsbug
flat current
Used the v19 tonight - seemed fine except: when in inventory or using a laser designator, frames dropped to 9-10 from 40-50. Came back up immediately upon exiting inventory or designator. Reverted to v18, no fps drop when using inventory or designator.
(using Performance, not profiling)
thick elk
I have a question: I used to manually extract the performance exe, now I just subbed to the profiling build(via steam beta), I see that it logs a little more, is there a performance difference, that you can actually tell?
silver roost
Alot of it, is crash fixes and optimisations, so depending on your system there can be alot to gain. Some have insane results, some have none or even opposit effect (rare). Atleast thats the experience with our community
thick elk
You mean the profiling exe runs better than the performance exe?
I am also a little confused about the last sentence of the post: "(performance binaries are used by default so nothing to worry about)"
I mean the normal Arma3.exe I get via Steam is not the performance.exe, I download from the dropbox, I am pretty sure....
silver roost
No profiling runs alot of debug, so that is NOT better performance. The performance from steam IS the same as the one from dropbox.
As per standard the profiling branch on steam runs the performance.exe
thick elk
Thanks for the last sentence
I was always confused, since it is called "profiling" I thought it downloaded the profiling one
So I mostly manually downloaded the perf. build
silver roost
Thats the name of the branch.
silver roost
๐
nocturne obsidian
on another note: i can confirm the low fps on map and inventory screen with the new v19
empty goblet
@old arrow the binary names determine usage, profiling .exe is with debug diagnostic layer, thus slower, while performance .exe is just like MAIN branch stable
also the names on manual download are different for that reason, must replace the existing arma3.exe or arma3server.exe , it's in the howto readme ๐
if you switch via STEAMclient you get the performance as automatic replace of .exes
using profiling binary has use only for those who tracking something on my/our request or want the debug script commands etc.
flat current
If I wanted to get the "diag captureframe snapshot" while using profiling, how would I do that? ๐ (Never used profiling)
figured it out
thick elk
@empty goblet Everything is very clear, the only thing that was not is, if I get the performance or the profiling .exe when I subbed to the profiling build via Steam. Imo it should just be called the performance build.
empty goblet
you get both, default arma3.exe / arma3server.exe are performance binaries types
it's profiling branch, it contains both performance and profiling binaries ... simple as that
boreal wigeon
My FPS goes to 6 when on map, camera and in inventory screen with V19.
woven loom
hmm that's interesting, i was having FPS issues on the map as well when at a certain zoom level with v18
nocturne obsidian
is there a way to disable the perf_prof watermark in the lower right but still use this branch?
nocturne obsidian
yes
queen owl
@nocturne obsidian try pressing pause key
empty goblet
nope, afaik dead end
empty goblet
1.66.140769 new PROFILING branch with PERFORMANCE binaries, v20, server and client, windows/linux
- fixed multiple Crashes (soldiers delete, broken skeletons, zeus unit grouping changes, wrong mission signatures etc.)
details:
https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3161970
nocturne obsidian
data verfication kick
empty goblet
? @nocturne obsidian ?
boreal wigeon
Does it fix the frames drops when in the inventory and such?
heady parcel
I dont want to jinx it, but I think my crashes around ai are gone...
fps in inventory and map do not drop for me on this build
nocturne obsidian
framedrops in inventory and map seem to occur when using ace and acre. not sure about vanilla. blaubaermitumlaut posted in the perf_prof thread in the forums why there is a FPS drop and asks kindly to look at this issue since the ace and acre mods are widely used by lots of people. since v19 brought us the fps drop, maybe there will be a fix in v21 for that, dwarden? ;D
empty goblet
i already told some that it's more likely related to the crash fix which was sorting some incorrect formats used, so if they doing that then it's more likely it
in such case i told them to contact me ideally with simple repro of what they do ... so far nobody returned to me with anything more
also it's best to test against latest DEV and RC too as if it's fixed there then it should not be much to worry
nocturne obsidian
thank you for your reply. i will talk to them and test v20 by myself ๐
heady parcel
I just got the same message after changing mission file. Non v20 clients seem to connect fine
stark falcon
Same thing, was on profiling exe, problem went away after reverting to normal.
21:52:14 Malformed data received. Setting default server data.
21:52:38 Warning Message: Data Verification Failed
Connecting to clean restated server was fine, only got kicked after selecting mission.
Server was not running profiling.
heady parcel
I switched back to vanilla was able to join but crashed after playing a few minutes. Switched back to performance re-verified and was able to connect again.
mossy flicker
Same over here. Data verfication failed. Both server and client running v20.
empty goblet
ok @nocturne obsidian @heady parcel @stark falcon @mossy flicker ... narrowed this to 3D editor missions guess we need revert of the sig crash fix until figured out
mossy flicker
I also tested to re-save my mission while in editor on v20, to see if it could fix it, but it didn't make any difference.
nocturne obsidian
regarding low fps in map and inventory screen with ace/acre: v20 has still fps issues. i talked to an ace dev and they will be in contact with you soon, dwarden. they mentioned this post in the forums in which they describe the reason of the fps issue. "baermitumlaut" said: https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3161290
lost wagon
@empty goblet I cant confirm the "Data verification failed" problem. All our servers are running with v20 - my client too. No problems so far ๐
empty goblet
@nocturne obsidian i need simple repro ... or whole script (e.g. github) which does it etc.
@lost wagon v21 enroute with revert
empty goblet
in meantime anyone got crash with 140769 ?
empty goblet
1.66.140795 new PROFILING branch with PERFORMANCE binaries, v21, server and client, windows/linux
- fixed more crashes
- fixed verification data issue on connect in v20
https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3162279
@nocturne obsidian @heady parcel @stark falcon @mossy flicker try new version , thanks
stark falcon
Connected fine with Version: 1.66.140795 (v21)
nocturne obsidian
testing next rs thanks
empty goblet
that ACE/ACRE related inventory fix next week ... (already done inhouse)
stark falcon
FWIW: next releases of ace and acre will have corrected use of locations to be like CBA, which does not have the issue
nocturne obsidian
outstanding news regarding ace/acre, dwarden! thank you for your hard work!
woven loom
hmm v21 didn't hold performance for me
last version that held 18-20 fps for me was v18
oblique panther
Anyone having problems with client crashing when trying to join Exile servers? I get Error Status Violation a lot. Drivers are up to date, settings is correct. I am on a GTX660/AMD FX8320 system and v19 and 20 gave me 50-60 FPS. I play on EXO servers only
I had least problems with V19, V20 generated more crashes and FPS drops, and now V21 or 1.66.140.795 crashes every time I try to join EXO's Exile Chernarus server. Mods are up to date.
empty goblet
v22 ready, has the fix for low fps issue vs ace/acre on map.
empty goblet
1.66.140847 new PROFILING branch with PERFORMANCE binaries, v22, server and client, windows/linux
- fixed performance drop at map/gps introduced in v19 ...
details : https://forums.bistudio.com/topic/160288-arma-3-stable-server-166-performance-binary-feedback/?do=findComment&comment=3163461
wanton grotto
So ran with v22 tonight and didn't get a 3fps crash, while normally on opnights i have had a few. Other changes are a windows reinstall, bios update and fresh download of all the game files. Will need to see if this was just a fluke or actually something that fixed the problem for me
i have a dxdiag and rpt's for before and after. no dumpfiles as i ended up alt+f4 instead of waiting for the process to fail :S (except for 1 old one from 4 weeks ago)
whole cloud
Will we have prof builds for 32 and 64 bit after 1.68 release? or only 64 bit?
nocturne obsidian
no more 32bit with 1.68