Dynamic campaign (alike) graphic test

Hello guys! I would like to share results of a test I made.

My goal was to test how much DCS engine can "take" managing air/land/sea units.

This are my specs:

Intel I5 7600 3.5GHz

SSD 256 GB

16 GB DDR3

MSI GTX 1060 6 GB

Monitor LG 34" 21:9 2560x1080

Windows 10 Pro

Track IR , NO VR

You can say it's an avarage gaming PC nowadays.

and settings:

https://scontent-mxp1-1.xx.fbcdn.net/v/t1.0-9/62055437_10216796490338675_2504039489177911296_o.jpg?_nc_cat=110&_nc_ht=scontent-mxp1-1.xx&oh=9dce920c20e5c73b798a99713fd22e84&oe=5D88C726

DCS version OB 2.5.5 , so with video memory leak fixed.

The mission is set in Caucasus , with a total of about 600 ground units , 100 aircrafts , 40 ships.

The map looks like that:

On the beginning of the mission , after click FLY , I get at least 60 sec of loading before actually having any response from the program. After that things starts.

I get constant 10 FPS while ships fire missiles against other ships and an amount of probably 50 tanks shoots each other very far from my position ( BATUMI airport ).

You can say it's a low intensity war at this point. Still more than 5 GB of video RAM is used .

The funny thing is, if I press ESC and pause the game I get 60 constant FPS.

https://scontent-mxp1-1.xx.fbcdn.net/v/t1.0-9/61950595_10216796492138720_5914182945983168512_o.jpg?_nc_cat=106&_nc_ht=scontent-mxp1-1.xx&oh=9b46f2a4b0e5ecfce1d800e7f95250c3&oe=5D8774A1

This is an unexpected behavior , at least for me , I thought the FPS I got was due to heavy graphic duty but this changed my mind.

During pause you GPU is still used but CPU is working less.

So , is the CPU the bottleneck? Probably!

I can totally see why ED is trying to implement Vulkan API and multi-thread programming.

You can basically sky rocket all the graphic and get very good results even with VR.

This is the mission if you want to test it too...

https://www.dropbox.com/s/11cvbfqp3sr5suu/TEST_DC.miz?dl=0

Would be amazing to see results with an high-end PC.

I'm not surprised at all. In most cases only a big high detail city or many high poly objects in the view can push your graphics geometry. Otoh you can kill the fillrate of any graphic card by setting high enough resolution (VR) and/or anti-aliasing setting. All the rest is CPU-limited - esp. the AI units.

So , is the CPU the bottleneck? Probably!

I can totally see why ED is trying to implement Vulkan API and multi-thread programming.

You can basically sky rocket all the graphic and get very good results even with VR.

your test is CPU limited, right?

doesn’t Vulkan offload the GPU and push the most of the “work” to the CPU?

how will giving the CPU even more things to do, speed things up?

how will giving the CPU even more things to do, speed things up?

It can assign the work to the currently bored cpu cores :)

your test is CPU limited, right?

 

doesn’t Vulkan offload the GPU and push the most of the “work” to the CPU?

 

how will giving the CPU even more things to do, speed things up?

It does not offload the GPU , it optimize the workload between GPU and CPU.

Look at this.

https://medium.com/@toncijukic/draw-calls-in-a-nutshell-597330a85381

In an engine with a lot of objects ( like a flight sim ) and drawcalls Vulkan "should" greatly improve FPS.

If I run this mission I'm getting max 60% CPU on any core and around 40% on my GPU. Still very low FPS though. Something else is going on here, I'm getting very similar results on a very simple mission with some AI and targets enabled. 5i 9600 k @ 5Ghz 32 meg RAM and an RTX 2080. I ran a perfmon trace on my issue and see no CPU, GPU or RAM readings that would explain it.

@beppe - I have an i9-9900k and 2080ti card. Will try the test tonight after work.

If I run this mission I'm getting max 60% CPU on any core and around 40% on my GPU. Still very low FPS though. Something else is going on here, I'm getting very similar results on a very simple mission with some AI and targets enabled. 5i 9600 k @ 5Ghz 32 meg RAM and an RTX 2080. I ran a perfmon trace on my issue and see no CPU, GPU or RAM readings that would explain it.

How many FPS you get? Are you using my settings?

No, everything maxed out at 4K. But as I said there is something wrong the performance of this release... 5-10 FPS when it's juddering 60 FPS in-between

25-35 fps with high/ultra settings and WQHD

the 1 core is full tilt at maximum, something is overloading the engine.

25-35 fps with high/ultra settings and WQHD

 

the 1 core is full tilt at maximum, something is overloading the engine.

:thumbup:

Pretty much every simulation in-game.

One thing I noticed is after 30 min , when ground units falls to 500 total a have about 20-25 FPS.

That maybe means ground units AI are CPU heavy.

Ok took a look at the mission. You can see my specs in my sig.

I found the following -

1. Using my VR headset, my frames never got above 24fps.

2. My GPU utilization stayed consistent at 34% - very low

3. My CPU never topped 25%

4. Only 1 core spiked above 40% for one millisecond. All cores averaged less than 10% utilization

5. Memory topped at 16GB used - the system has 32GB available.

Mission was totally unplayable. FPS ranged from 10-24 FPS and was very jerky in the HMD. Yet none of my resources in MSI were tapped.

Ok took a look at the mission. You can see my specs in my sig.

Mission was totally unplayable. FPS ranged from 10-24 FPS and was very jerky in the HMD. Yet none of my resources in MSI were tapped.

 

I can see exactly the same issue not using VR, no resources maxed out, FPS drops to below 10 as soon as the AI engages the SAM site in my mission, reproducible every time using this track file

https://forums.eagle.ru/attachment.php?attachmentid=211692&d=1560171327

I can see exactly the same issue not using VR, no resources maxed out, FPS drops to below 10 as soon as the AI engages the SAM site in my mission

Well, with bug like this and mission that big and complex, it's mission impossible to find the problem. You'd have to start over and consistently check the performance after adding groups of units.

Well, with bug like this and mission that big and complex, it's mission impossible to find the problem. You'd have to start over and consistently check the performance after adding groups of units.

If you play the track file you'll notice that there are only a few groups of AI aircraft and one SAM group... The other AI (AWACS etc) has no bearing on the problem, removal of the SU34 group or the SAM group means the problem doesn't show up. I've already raised a bug with the trk file, log, and a data collector set.

https://forums.eagle.ru/showthread.php?t=242962

If you play the track file you'll notice that there are only a few groups of AI aircraft and one SAM group... The other AI (AWACS etc) has no bearing on the problem, removal of the SU34 group or the SAM group means the problem doesn't show up. I've already raised a bug with the trk file, log, and a data collector set.

I know - I was talking about beppe's misson :) So it may or may not be only the quantity problem there.

DCS performance is severely affected by the number of AI units.

The funny thing is, if I press ESC and pause the game I get 60 constant FPS.

https://scontent-mxp1-1.xx.fbcdn.net/v/t1.0-9/61950595_10216796492138720_5914182945983168512_o.jpg?_nc_cat=106&_nc_ht=scontent-mxp1-1.xx&oh=9b46f2a4b0e5ecfce1d800e7f95250c3&oe=5D8774A1

 

This is an unexpected behavior , at least for me , I thought the FPS I got was due to heavy graphic duty but this changed my mind.

During pause you GPU is still used but CPU is working less.

That's probably because DCS processes were paused.

Even though apparently recent udates have helped with alot of units, especially in MP. It all grinds to a halt, if those units do anything other than sit static waiting to be blown up. Hence most missions you find just have static targets everywhere. The fact ED have announced this "dynamic campaign", i think will come back to bite. I dont see with the current engine even half of a fully dynamic conflict being playable.

I hope i'm wrong and we see massive strides in performance in the near future, but history with ED would suggest not.

So at the present, does DCS use the concept that most should know about, "Bubble". This helps with frame rate problems. That's the main reason why (un-named sim) can run a full time war in the background and can be run with perfectly acceptable frame rates.

I don't get why you guys are on one core, to check I just took a pic of the processes during the P51 Instant Action Ground attack mission as I flew over AI while being fired on. Affinity is checked off on all cores.. Of course it doesn't matter with the test mission, 10-15FPS until paused..

I don't get why you guys are on one core, to check I just took a pic of the processes during the P51 Instant Action Ground attack mission as I flew over AI while being fired on. Affinity is checked off on all cores.. Of course it doesn't matter with the test mission, 10-15FPS until paused..

You are mislead by the graph.

Threads jump across all your cores and depending on your monitoring resolution ( 1ms, 0.5sec, 1sec etc. ) this does look like it is using ALL cores, but it is NOT.

Follow the Hardware section part of this forum for more insight in this. It is a very complicated matter and different CPU's and OS versions behave differently, which doesnt make it any easier to tell what is actually happening with the DCS threads.

:thumbup:

 

Pretty much every simulation in-game.

 

One thing I noticed is after 30 min , when ground units falls to 500 total a have about 20-25 FPS.

That maybe means ground units AI are CPU heavy.

To be more specific: Moving AI units are CPU heavy. This is well known and the reason why multiplayer servers (especially the ones that are running big dynamic campaigns like Blue Flag) keep their AI ground units static.

Here are some comments from Ciribob (one of the main guys behind Blue Flag and also a member of the ED testers team on this:

https://forums.eagle.ru/showpost.php?p=3563327&postcount=141

https://forums.eagle.ru/showpost.php?p=3160055&postcount=2164

Other games like DCS with dynamic campaigns do solve this by using the mentioned "bubble technology". With this only units in a certain radius around players (say 50km for example) will actually be fully simulated. All units outside this bubble and their actions will only be computed on an abstract level which takes much less CPU performance.

Here's a nice interview with someone who developed the most well known dynamic campaign for a flight sim and he talsk how difficult it is to get a lot of AI units together with low CPU requirements: http://web.archive.org/web/20110318015840/http://www.cleared-to-engage.com/2011/03/12/interview-with-kevin-klemmick-lead-software-engineer-for-falcon-4-0

You are mislead by the graph.

 

Threads jump across all your cores and depending on your monitoring resolution ( 1ms, 0.5sec, 1sec etc. ) this does look like it is using ALL cores, but it is NOT.

 

Follow the Hardware section part of this forum for more insight in this. It is a very complicated matter and different CPU's and OS versions behave differently, which doesn't make it any easier to tell what is actually happening with the DCS threads.

Thanks, i'm easily misled.. :thumbup:

My latest venture has been using HWiNFO64 to monitor CPU and GPU temps on this laptop. It looks like the CPU Cores 1,2,3 are Thermal throttling down and the CPU drops from 40x (100mz) to 32x. Funny thing is the FR in game never drops below 140+ while viewing.. Temps are a bit high though, almost 100C across all cores. Idle is at 53C.

Edit: Killed the OC / lowered the GPU to 60Hz and all temps dropped substantially..