fixed Huge SP performance degradation with 2.9.7.58923 and 2.9.6.58056 compared to DCS 2.9.5.55918

[The_Nephilim]

Hey Guys, I retried to do the Scenic Bench mission and I found out you need the F-18 and SuperCarrier which I have will report back.

 

Well Last night when I tested it I did not have the SC installed and when I was looking at the Su-25's on the ground the FPS counter was flatlined.. I just installed the SC and when I was on the deck it was holding perfect FT's but the FPS were jumping all over the place..

Last night I had OC my RAM to 4600MT's and that is when looking at the Su-25's it seemed smooth but maybe the SC is inducing the bad FPS?? I have read about a fix for the SC anybody know what it is?

Edited September 8, 2024 by The_Nephilim

[Rene Coulon]

10 hours ago, Pande4360 said:

2.9.5.55918 works fine for me

I wish I had stayed there !

[Darcaem]

Just in case anyone want to try, I have experienced another performance gain (this is by no means a fix, but at least performance is "not as bad") rescaling all textures using Zbysiek script

 

I've just played a pretty rutine and boring CAS retribution mission on the A10CII in Sinai with no stutters!! And it has been the most awesome flight I've had in months in DCS XD

 

To summarize, tricks that had improved performance for my (in VR with i9700k, 64GB@3200, rtx3700, 2TB m.2). Some might perfectly be placebo and I don't really know which one actually do have any effect.

* 64GB static pagefile on the same m.2 drive that DCS (OS installed on a regular SSD)

* Fast VSync on NVidia control panel

* Rescaled textures

* Affinity core trick

* Obviously reducing the number of units on my custom missions and remove OH6A gunner script...

null

[Lange_666]

Triple buffering set to ON?

If you hoover over it, it says: Allows you to enable or disable triple buffering for "OpenGL" applications.

[Darcaem]

Yep, I know, totally stupid. However, after many years I have taught myself not to defy the Computers God's placebo effect It has absolutetly no effect, but I'm following the "if aint broke, don't fix it" rule. The same with the fast vsync value, since vsync is off on DCS it probably has no actual effect...

[Jara4]

On 9/8/2024 at 12:20 PM, Rene Coulon said:

I wish I had stayed there !

You can downgrade. I did and it plays well again.

Edited September 9, 2024 by Jara4

[Stealth690]

Hello,

And another .log, if this can help

The stutters only started at the point where multiple AI aircrafts are introduced. This is a singleplayer campaign mission from Operation Cerberus North (M13 Part2).

 

Thanks,

Cheers

dcs.log

[metzger]

I think I face the same issue. On missions starting on an airbase with multiple other AI aircraft and ground units, first the mission when you have the window to press 'fly' loads the textures very slowly, like 10 minutes or so it can't load the cockpit textures. If I alt+tab I can see 100% ram is used.

Once the textrures and objects are finally loaded, click fly and it is a slide show, sometimes it might get smooth for 10 seconds and starts stuttering heavily again with single core CPU loaded and all the ram utilized.

My system:

Ryzen 7 5800X3D

32GB ram

RTX4080

Win 10

Quest3

[Pande4360]

Am 12.9.2024 um 12:41 schrieb metzger:

I think I face the same issue. On missions starting on an airbase with multiple other AI aircraft and ground units, first the mission when you have the window to press 'fly' loads the textures very slowly, like 10 minutes or so it can't load the cockpit textures. If I alt+tab I can see 100% ram is used.

Once the textrures and objects are finally loaded, click fly and it is a slide show, sometimes it might get smooth for 10 seconds and starts stuttering heavily again with single core CPU loaded and all the ram utilized.

My system:

Ryzen 7 5800X3D

32GB ram

RTX4080

Win 10

Quest3

Try the same mission but start far away on an empty airfield then you can see the 2 issues seperated.

[buceador]

Something more to add to the pot of stutters and maybe this is already known...

I created a ST mission, comparatively simple, I deliberately limited the amount of ground units and used statics wherever possible.  As I taxied out in the F-16 the new CH47F was triggered to take off from ground hot about 200 metres from the Viper, as soon as it spawned and started to move the stutters started, so badly that I exited and replaced the new Hook with the standard DCS version.  Restarted the mission and it was totally smooth. 

[Darcaem]

45 minutes ago, buceador said:

Something more to add to the pot of stutters and maybe this is already known...

I created a ST mission, comparatively simple, I deliberately limited the amount of ground units and used statics wherever possible.  As I taxied out in the F-16 the new CH47F was triggered to take off from ground hot about 200 metres from the Viper, as soon as it spawned and started to move the stutters started, so badly that I exited and replaced the new Hook with the standard DCS version.  Restarted the mission and it was totally smooth. 

I think that is directly related to CH47F textures and vram usage. Correct me if I'm wrong, but I believe we are facing two different problems that are intertwined: newer modules with huge unoptimized texture file sizes (while at the same time older modules have smaller textures, so if you decrease texture settings newer cockpits look fine but older ones are unreadable), and some changes in ground unit pathfinding and/or detection having a much greater impact (so DCS is now able to handle way fewer units than before)

EDIT: have you tried this?

 it has improved my performance with zero visual quality lost

Edited September 16, 2024 by Darcaem

[Canada_Moose]

These huge texture files are completely unnecessary. The well known civilian sim has been using compressed high res textures for years with no issues.

Jabbers recent video regarding VRAM usage in DCS clearly shows how these huge model textures are impacting things. Its especially annoying when these same flyable model textures are being loaded in for AI use.

[bojote]

Fresh install (DCS installed for the first time), no mods, Just Hornet, Syria and Cerberus North. Mission 16 (Part 2) 5 seconds after pressing spacebar:

DCS 2.9.5.55918

vs

DCS 2.9.6.58056

Needless to say, AFTER 2.9.5.55918 something messed up heavy unit SP missions. Fact, easy to replicate.

Edited September 18, 2024 by bojote

[bojote]

On 9/3/2024 at 5:07 PM, SharpeXB said:

I don’t mean to derail the thread, these things are important to look into. Sometimes it might help to know who isn’t affected as well as who is. 

I believe thats precisely the point everyone here is trying to make. Its not a matter of 'who' is affected, but when will it affect you. If you only fly single player instant missions or simple MP dogfight PvP servers then you'll probably NEVER be affected and you'll post things such as: Everything fine on my end! no performance issues here.. thats fine! but, what is being discussed here is something much more complex. Its a LOGIC/PROGRAMMING FLAW in the way AI Ground and Air units interact. As if an endless loop and/or unnecesary recurring process is trashing the CPU, most likely due to a race condition and/or a scoping issue (my theory is that ALL ground units and actively searching for AI air threats, instead of a selected few within a zone/bubble)

The reason you see people frustrated is because they know that 2.9.5.x did NOT have this issue, period. This thread and the problem discussed has been already identified and being worked on by ED, my concern (and this is nothing against Scott) but something that worries me is when he says ED is working on 'Optimizations' and finding ways to make FPS better. We don't need an 'optimization'. This is a BUG, a problem in the code, a LOGIC issue and the way to fix it is to either ROLL back the AI Ground/AIR interaction logic introduced in 2.9.6.x or rework its implementation as its clearly not working. 

[SharpeXB]

13 minutes ago, bojote said:

If you only fly single player instant missions or simple MP dogfight PvP servers then you'll probably NEVER be affected

I did actually test the OPs benchmark miz and had no trouble with it. I’m not denying there’s a problem somewhere just that I couldn’t replicate it. 

[bojote]

7 minutes ago, SharpeXB said:

I did actually test the OPs benchmark miz and had no trouble with it. I’m not denying there’s a problem somewhere just that I couldn’t replicate it. 

Do you own Operation Cerberus North? A consistent way to test and see the problem inmediatly is to load Mission 16 (Part 2).. that SAME mission in 2.9.5 works FLAWLESLY, not a single stutter and consistent 250-280 fps. If you load the exact same mission with > 2.9.6.x you'll inmediatly spot the problem, FPS will wildy go from 40-150 and frame time (which is the time it takes to render a single frame) will go up and down erratically.

If you edit that mission and remove either ALL AIR AI units or GROUND/STATIC units (there are > 2000 units aprox) then everything works fine. The attached image shows what changed after 2.9.5.x. Thats why you see people suggesting that the problem has some relation to the logic/interactions between Air AI and Ground units. ED also MOVED the logic to a different thread, so there is a thread syncronization matter that could potentially affect the MAIN thread. If the main thread is 'expecting' something to happen and/or needs to wait for it you'll have a condition in which the CPU will consume cycles, go 100% and stall the entire simulation. As discusses before, this is a complex programming issue that requires specific conditions in order to be triggered and that condition is simply lots of units (not sure if ground, static, blue or red) thats for ED to investigate (which they are)

null

[BIGNEWY]

1 minute ago, bojote said:

Do you own Operation Cerberus North? A consistent way to test and see the problem inmediatly is to load Mission 16 (Part 2).. that SAME mission in 2.9.5 works FLAWLESLY, not a single stutter and consistent 250-280 fps. If you load the exact same mission with > 2.9.6.x you'll inmediatly spot the problem, FPS will wildy go from 40-150 and frame time (which is the time it takes to render a single frame) will go up and down erratically.

If you edit that mission and remove either ALL AIR AI units or GROUND/STATIC units (there are > 2000 units aprox) then everything works fine. The attached image shows what changed after 2.9.5.x. Thats why you see people suggesting that the problem has some relation to the logic/interactions between Air AI and Ground units. ED also MOVED the logic to a different thread, so there is a thread syncronization matter that could potentially affect the MAIN thread. If the main thread is 'expecting' something to happen and/or needs to wait for it you'll have a condition in which the CPU will consume cycles, go 100% and stall the entire simulation. As discusses before, this is a complex programming issue that requires specific conditions in order to be triggered and that condition is simply lots of units (not sure if ground, static, blue or red) thats for ED to investigate (which they are)

null

Can I ask you to check something, do you use tacview? if so please remove the scripts folder from saved games dcs and retest. Let me know. 

thank you 

[Hiob]

5 minutes ago, bojote said:

Do you own Operation Cerberus North? A consistent way to test and see the problem inmediatly is to load Mission 16 (Part 2).. that SAME mission in 2.9.5 works FLAWLESLY, not a single stutter and consistent 250-280 fps. If you load the exact same mission with > 2.9.6.x you'll inmediatly spot the problem, FPS will wildy go from 40-150 and frame time (which is the time it takes to render a single frame) will go up and down erratically.

Actually DCS has a habit of freaking out when the FPS approach 300. Try to limit the FPS to 180 and see if that helps with consistent frametimes. Maybe you simply got better fps with the update and hit the treshold.

Edited September 20, 2024 by Hiob

[bojote]

3 minutes ago, Hiob said:

Actually DCS has a habit of freaking out when the FPS approach 300. Try to limit the FPS to 180 and see if that helps with consistent frametimes. Maybe you simply got better fps with the update and hit the treshold.

 

Its limited to 120. The 250-280 number is the theoretical you see in the frame counter and simply used for reference

[SharpeXB]

18 minutes ago, bojote said:

Do you own Operation Cerberus North? A consistent way to test and see the problem inmediatly is to load Mission 16 (Part 2).. that SAME mission in 2.9.5 works FLAWLESLY

I think I do but I haven’t got to it yet. I don’t have 2.9.5 to compare with either. Your report clearly identifies a problem though. 

[Hiob]

30 minutes ago, bojote said:

Its limited to 120. The 250-280 number is the theoretical you see in the frame counter and simply used for reference

roger

[bojote]

13 minutes ago, BIGNEWY said:

Can I ask you to check something, do you use tacview? if so please remove the scripts folder from saved games dcs and retest. Let me know. 

thank you 

I do in my main DCS computer, but since troubleshooting this problem I installed DCS in my (other civilian sim) computer, which NEVER had DCS installed. Meaning no mods AT ALL. Zero. Only Hornet, Cerberus North and Syria. Thats it.

As I said, I've tried (on my own) to isolate the problem (as others here are doing) and what I did, was to roll back my DCS version to 2.9.5.x. which works great. My next step is using process monitor (the sysinternals tool) and compare what DCS.exe is doing in 2.9.5.x different from > 2.9.6.x that might help. Also, if anyone is familiar with process monitor (NOT THE WINDOWS process monitor) but the specialized tool you download you'll see how helpful that can be. I don't thinking bashing ED will help, thats why I rather take a proactive approach and see if I can spot something obvious that will help the devs figure out what the problem is.

 

[Pande4360]

All you need is to run a liberation mission

Here´s an example whcih i just tested. Not the worst mission but you can see some differences. Smooth 72fps (capped) in 2.9.5 and a lot of dips in 2.9.7 .

Some other interesting things in 2.9.7 checking the cores one core was reaching the 100 frequently. After disabling the core affinity of that core and another spiky one things spread out more and there was slightly less stuttering, though still there to some degree.

retribution_nextturn.miz

Edited September 24, 2024 by Pande4360

[Rene Coulon]

On 9/20/2024 at 2:34 PM, bojote said:

I believe thats precisely the point everyone here is trying to make. Its not a matter of 'who' is affected, but when will it affect you. If you only fly single player instant missions or simple MP dogfight PvP servers then you'll probably NEVER be affected and you'll post things such as: Everything fine on my end! no performance issues here.. thats fine! but, what is being discussed here is something much more complex. Its a LOGIC/PROGRAMMING FLAW in the way AI Ground and Air units interact. As if an endless loop and/or unnecesary recurring process is trashing the CPU, most likely due to a race condition and/or a scoping issue (my theory is that ALL ground units and actively searching for AI air threats, instead of a selected few within a zone/bubble)

The reason you see people frustrated is because they know that 2.9.5.x did NOT have this issue, period. This thread and the problem discussed has been already identified and being worked on by ED, my concern (and this is nothing against Scott) but something that worries me is when he says ED is working on 'Optimizations' and finding ways to make FPS better. We don't need an 'optimization'. This is a BUG, a problem in the code, a LOGIC issue and the way to fix it is to either ROLL back the AI Ground/AIR interaction logic introduced in 2.9.6.x or rework its implementation as its clearly not working. 

I agree COMPLETELY

On 9/16/2024 at 7:33 PM, Darcaem said:

I think that is directly related to CH47F textures and vram usage. Correct me if I'm wrong, but I believe we are facing two different problems that are intertwined: newer modules with huge unoptimized texture file sizes (while at the same time older modules have smaller textures, so if you decrease texture settings newer cockpits look fine but older ones are unreadable), and some changes in ground unit pathfinding and/or detection having a much greater impact (so DCS is now able to handle way fewer units than before)

EDIT: have you tried this?

 it has improved my performance with zero visual quality lost

 

How do we use this script, extract it and where must we load it ?

[Lange_666]

1 hour ago, Rene Coulon said:

How do we use this script, extract it and where must we load it ?

Did you open the link and read what you have to do?
It's all explained in steps.