Possible fix for the "3 FPS Drop" Issue

[kridian 33]

Yeah, this new one actually recovered from the glitch 3 times and then couldn't (while being air lifted out of an AO). Performance was choppy during the peak.

[bis_iceman 7384]

Thanks for all the responses so far, I'm tracking them closely. Let's focus on the original Allocator first (in the starting post), please. We can decide later where to move with the research. Your help is very appreciated!  :)

[ruPal 143]

As you work on memory allocator are there any plans to implement Fred41 tweaks?

[Plainfield 2]

So far its working well, I slightly have a lower fps than I did before but that could be on video settings. The *texture dump* did happen once so far but recovered, also that was in a 110 man server; keep up the great work!

 

gigabyte z97x-ud3h-cf

4790k @ 4.8

Nvidia Zotac AMP Extreme 1080 - 3 screens

16gb corsair vengeance @2400
SSD pro 950

EDIT: still having isssues

[Dwarden 1125]

As you work on memory allocator are there any plans to implement Fred41 tweaks?

note: Fred41 never published changes from the latest reivisions of private allocator as public code ...

note: large pages support may come but so far no eta ...

[BountyHuntA 10]

I used plain main branch and also main + tbb4 dll from 10th october from this source: https://forums.bistudio.com/topic/160288-arma-3-stable-server-164-performance-binary-feedback/page-86

 

I still can experience the 5fps bug. I can relatively fast provoke it with this:

In a busy mission with lots of AI fly around with admin spectator cam (splendid cam?). Do jumps via clicks on the spectator map. Also turn spectator cam off and on multiple times.

[SteemSucker 19]

So far working for me, will report back if I experience any crashes. --I will do an fps comparison with the old allocator as well as try this on my desktop system. 

 

Laptop:

Win10 64bit

i7 6820HK @ 4.0

32GB RAM

GTX 1070 8GB

 

Desktop:

Win10 64bit

i7 6700K @ 4.6

32GB RAM

GTX 780 3GB (No h8 pls im saving for a 1080  :P )

 

***********

UPDATE

***********

Laptop: After 1.5 hours of editing a mission it was still stable and running well. Did not notice a large difference in FPS however I did not have a chance to "benchmark" it yet. 

 

 

Desktop: After about 4 hours on multiplayer I had the 3 FPS issue immediately followed by a crash; "DX11 error: CreateTexture failed : E_OUTOFMEMORY" and "The exception unknown software exception (0x0000dead) occured in the application at location 0x000000007762DB18." 

Attached is the full troubleshooting report but I don't know if that's helpful. 

 

Note, I was using the main branch with the tbb4 allocator on both these systems. I will be trying it with the profiling branch tomorrow. 

[das attorney 858]

:(

 

10:51:09 VM busy 4107096064 B (reserved 205922304 B, committed 3901173760 B, mapped 177782784 B), free 187740160 B
10:51:09 Small mapped regions: 25, size 118784 B
10:51:10 Virtual memory total 4095 MB(0 B)
10:51:10 Virtual memory free 195 MB(0 B)
10:51:10 Physical memory free 6216 MB(0 B)
10:51:10 Page file free 5279 MB(0 B)
10:51:10 Process working set 2943 MB(3086733312 B)
10:51:10 Process page file used 3457 MB(3625377792 B)
10:51:10 Longest free VM region: 14594048 B
10:51:10 VM busy 4107161600 B (reserved 205836288 B, committed 3901325312 B, mapped 177848320 B), free 187674624 B
10:51:10 Small mapped regions: 25, size 118784 B
10:51:10 Virtual memory total 4095 MB(0 B)
10:51:10 Virtual memory free 195 MB(0 B)
10:51:10 Physical memory free 6197 MB(0 B)
10:51:10 Page file free 5279 MB(0 B)
10:51:10 Process working set 2943 MB(3086950400 B)
10:51:10 Process page file used 3457 MB(3625406464 B)
10:51:10 Longest free VM region: 14594048 B
10:51:10 VM busy 4107161600 B (reserved 205807616 B, committed 3901353984 B, mapped 177848320 B), free 187674624 B
10:51:10 Small mapped regions: 25, size 118784 B
10:51:10 DX11 error : CreateTexture failed : E_OUTOFMEMORY
10:51:11 DX11 error : CreateTexture failed : E_OUTOFMEMORY
10:51:11 DX11 error : CreateTexture failed : E_OUTOFMEMORY
10:51:12 DX11 error : CreateTexture failed : E_OUTOFMEMORY
10:51:12 DX11 error : CreateTexture failed : E_OUTOFMEMORY
10:51:13 DX11 error : CreateTexture failed : E_OUTOFMEMORY
10:51:13 DX11 error : CreateTexture failed : E_OUTOFMEMORY
10:51:13 DX11 error : CreateTexture failed : E_OUTOFMEMORY
10:51:13 DX11 error : CreateTexture failed : E_OUTOFMEMORY
10:51:13 DX11 error : CreateTexture failed : E_OUTOFMEMORY
10:51:13 CreateTexture failed : w = 4096, h = 4096, format = BC3_UNORM, err = E_OUTOFMEMORY.
10:51:13 DX11 error : CreateTexture failed : E_OUTOFMEMORY
10:51:13 Virtual memory total 4095 MB(0 B)
10:51:13 Virtual memory free 195 MB(0 B)
10:51:13 Physical memory free 6198 MB(0 B)
10:51:13 Page file free 5279 MB(0 B)
10:51:13 Process working set 2944 MB(3087089664 B)
10:51:13 Process page file used 3457 MB(3625488384 B)
10:51:13 Longest free VM region: 14594048 B
10:51:13 VM busy 4107161600 B (reserved 205725696 B, committed 3901435904 B, mapped 177848320 B), free 187674624 B
10:51:13 Small mapped regions: 25, size 118784 B
10:51:13 VID: dedicated: 3221225472, shared 1073676288, system: 0, max: 2906652672, used: 2364776448
ErrorMessage: DX11 error : CreateTexture failed : E_OUTOFMEMORY

 

[fn_Quiksilver 1636]

Excellent, thanks for your time spent looking into it and working on a solution. I'll report back in a few days.

 

Reporting back, issue still persists.

[SteemSucker 19]

The profiling branch produces the same results, if not slightly worse. About every three to four hours on multiplayer I get a crash.

[luetin09 15]

Reporting in on this, it still persists unfortunately.

 

on my livestream this week I had two crashes and generally abnormal frame drops. One I was able to salvage with a quick game restart the other black screened me and effectively ended our gamenight.

As other people have stated it seems that generally having a lot of action on screen is what provokes it. I generally am in the spectator camera as Zeus more than other people and this could explain why I get the fps bug and consequent memory crash as I have more on screen both the AI icons object icons rendering landscape at a greater distance and so on. Its almost a given that if I am observing an intense situation I am going to get the fps drop hard.

 

It also seems to happen just as action kicks off, essentially it seems that when you are in a situation where the game will require more memory instead of handling this it seems to spiral out of control exponentially

 

So its still there :mellow:

[tilahunted 2]

I have tried overclocking my memory, removing and adding memory, overclocking/downclocking the gpu and cpu, and tried downloading the game onto the HDD and SDD. Still nothin'. I reckon it might help to go back to an old allocator until a permanent fix can be found? I don't know what Arma 2 used but that game ran smooth as honey whiskey!

[cyruz 103]

I ran in to the same 3fps issue with the new malloc, interesting side note, when playing I'm always recording with Dxtory, when this bug occurs, if I stop recording it buys me around another 30 minutes before the inevitable crash comes. Not sure what to make of this, I'm sure better minds might find it helpful however.

[HeroesandvillainsOS 1504]

Why isn't anyone posting rpts/logs/crash dumps? I don't see what good it does to come all the way in here and say you have the issue but not just go that really easy extra step and post the files.

Do they not need these things?

I'd help out in a second but it's not happening to me (fortunately).

[CallMehTOMMEH 126]

All recorded after the implementation of the new 2017 memory allocator and performance binary. Still getting the notorious 3 FPS bug as shown in this vid:

 

https://www.youtube.com/watch?v=3TBwn8tHFA0

[.kju 3244]

Is BI not providing extra debug builds with additional logging???

[KakalacKidA 1]

Followup to my first post - got the 3fps bug after 4 to 5 hours.. still seems more stable than before. Not a solution but a promising development! Thanks guys!

 

EDIT: Alt-Tabbed to write this - went back to game and it recovered!

 

EDIT #2: KOTH round finished, tried to load next mission and Arma crashed with Unable_to_init_DXGI. Had to Winkey+tab to bring up task manager in a separate desktop session to quit.

[CNutter 24]

Still getting the bug, all hopes lost. It was definitely cured for me with the the first version of the new malloc (dated 11 Oct iirc) and the previous profiling branch (1.64.138690). That combo if binary and malloc fixed the bug 95% for me.

The newer profiling builds, main branch, and using the new malloc with them has resulted in 3FPS bug crashes within <2 hours of playtime (sometimes less than 5 minutes).
 

Why isn't anyone posting rpts/logs/crash dumps? I don't see what good it does to come all the way in here and say you have the issue but not just go that really easy extra step and post the files.

Do they not need these things?

I'd help out in a second but it's not happening to me (fortunately).

I'd say post them anyway, it can't hurt!

 

Is BI not providing extra debug builds with additional logging???

You could check out Dwarden's profiling builds thread.

[.kju 3244]

@ CNutter

 

The point was about the average user/tester here - those won't use these separate builds to test and provide logs.

[elephant 12]

So with the most recent patch I've gained a few FPS- running at about 7-10 rather than 4-6.  However it's pretty clear the problem is persisting; I should have far better and changing settings does nothing. Can we expect another attempt at fixing this?

[DanielNL 3]

Nope didn't fix it, still happening on i7 2600k 4.5 GHz, 2x8 + 2x2 GB DDR3 RAM, GTX 960 4GB GDDR. (seems this issue only occurs when flying around the map a long time)

[DanielNL 3]

Btw happening on max viewdistance setting (this seems to be the trigger)

[Plainfield 2]

Sry for the post, I am trying to figure out how to upload my reports to this section :(

[serjames 357]

Any update on when this might get fixed to Main Branch.. ?

 

We had a cracking mission ruined tonight by 75% of players getting FPS reduced basically to 2-3FPS rendering the entire troop combat ineffective.... :-(

[.kju 3244]

It is not fixed - not in dev nor in performance branch. BI still has to do more work.