Memory related stutter help

[bandit648]

I've been troubleshooting an issue I have with intermittent stutter for months now. Generally I see a quick split second drop down to 30-40 fps and then back to 60. This can happen on an empty map with no units present.

 

At first I thought it was related to static clouds but I've later discovered that the clouds just make it worse and are not the cause. While running my test track I setup perfmon to monitor my disk and memory activity and I found that the exact same time I get the frametime spike I get disk and memory read spikes as well.

 

Apparently DCS is choking while trying to read terrain information as I move through the virtual world. I was able to prove this with a simple test. I setup my test track that is just floating at 400 knots over the ocean towards the terrain off in the distance. I get frametime spike and stutter at the exact same time, every time. If I run the same mission but without moving and just holding at 0 knots I get no stutter. The game is stuttering when trying to read terrain information ahead from disk and memory. Cycling F11 view makes it better but doesn't completely eliminate the issue.

 

I've tested my memory for 5 hours of memtest86 v8 and no errors. This may just be a DCS engine issue but I'd sure like to know if it's something on my end that can be fixed. Does anyone have any ideas?

 

This is my test track floating towards the terrain @ 400 knots

 

 

This is the same track standing still over the ocean

 

 

I've attached the track I've used for testing. If someone who knows how to use MSI Afterburner could run this track and record the results I'd really appreciate it. For me, the track stutters with frametime spikes every time at about 20 seconds in.

CA_benchmark_clouds.trk

[BitMaster]

I will give it a shot tonight and report back.

[bandit648]

I will give it a shot tonight and report back.

 

Thank you sir!

[BitMaster]

On my end it goes really well. The only thing you see is when I switch view to F2 at the beginning.

 

I had it run several times, everytime like this.

 

* 2nd pic shows fps and frametime, no drops.

Edited March 14, 2019 by BitMaster

[bandit648]

On my end it goes really well. The only thing you see is when I switch view to F2 at the beginning.

 

I had it run several times, everytime like this.

 

* 2nd pic shows fps and frametime, no drops.

 

Don't hit F2 after starting the replay track. The camera will just float over the clouds. Also, can you run it capped at 60 fps? If you setup a logging toggle key on afterburner you can start the log as soon the track starts to eliminate all the spikey stuff from loading.

 

That will clean up the graph. If you don't have any stutter then I might have a memory issue I need to look into more.

 

Thanks

[BitMaster]

Will do asap

 

 

Gesendet von iPhone mit Tapatalk

[BitMaster]

solid 60 from A - Z, locked at 60 and started 1 sec after track starts, tho it made no difference in the ugly graph :huh:

[bandit648]

solid 60 from A - Z, locked at 60 and started 1 sec after track starts, tho it made no difference in the ugly graph :huh:

 

 

Thanks so much for helping me with this. A couple of questions. Isn't the dip in framerate I circled a stutter? That looks like shortly after you started the track. Mine normally stutters like that 20 seconds in.

 

Also, it looks like your frametime is 0ms?? That is odd.

[BitMaster]

no, that is the moment I hit ALT-GR + F11 shortly after starting the replay to start the recording/logging.

 

For me, assigning a keymap or not to start logging does not change the graphs as I thought it would after you told me. It always graphs, always, from the moment MSI Afterburner starts until I log out or quit the app. So that much to clearing them up, hmmm.

 

There are no drops in either of the 2 screenshots. So let us conclude it must be something unique on your end that is not common with others.

 

You could open Ressource Manager and see what fires up when that happens. Might give a hint.

 

But any measuring itself causes interference, so this is always a 2 edged sword and you must distinguish between induced and actual behaviour. Not easy.

 

Could be some very stupid thing. i would start by taking everything OFF that is not needed.

I mean, quit apps, unplug unneeded USB, turn LAN printers off, TV LAN off, stuff like that has been found to cause such things, very odd but true.

 

BTW, update your drivers to 1809 compatible versions. Your Asus shares many drivers with mine, read the info given with the drivers from 1803 and later, Maybe your origin is buried there. I had to reinstall along that way as well. Its almost impossible to follow the advice without doing it from scratch with 1809 drivers, or 1803 back then.

 

Last but not least, the dip was no stutter, actually I hit the keys at that moment and didnt really focus on stutter in that specific splitt second, but I am pretty sure it was fluid as I had my son watching it and I told him what we did. So he would have said, "DAD, there is the stutter"..but he didnt.

 

 

I occasionally get stutter, or better delay, in MP, but that is due to MP's nature and not a client side issue.

[bandit648]

no, that is the moment I hit ALT-GR + F11 shortly after starting the replay to start the recording/logging.

 

For me, assigning a keymap or not to start logging does not change the graphs as I thought it would after you told me. It always graphs, always, from the moment MSI Afterburner starts until I log out or quit the app. So that much to clearing them up, hmmm.

 

It looks like you captured screenshots of the real-time graph. When you enable logging the selected graph data is saved to a log file in the configured path. You can then open/view the graph with Afterburner. See my attached screen for that. Also, we only need to select GPU Usage, Memory Usage, RAM usage, Framerate and Frametime. You can select those in the Monitoring tab of MSI Afterburner settings.

 

I also like to select those metrics in the OSD. You can select "text,graph" for the frametime and this will allow you to visually see the spike/stutter when it happens.

 

You could open Ressource Manager and see what fires up when that happens. Might give a hint.

 

I've been through everything trying to nail this down. Resource Manager, Performance Monitor, MSI Afterburner... The only thing that happens at the same time is disk read and memory read spike.

 

Could be some very stupid thing. i would start by taking everything OFF that is not needed. I mean, quit apps, unplug unneeded USB, turn LAN printers off, TV LAN off, stuff like that has been found to cause such things, very odd but true.

 

BTW, update your drivers to 1809 compatible versions. Your Asus shares many drivers with mine, read the info given with the drivers from 1803 and later, Maybe your origin is buried there. I had to reinstall along that way as well. Its almost impossible to follow the advice without doing it from scratch with 1809 drivers, or 1803 back then.

 

I haven't tried disconnecting devices so I will give it shot. I'll disconnect everything I can and then re-run the test. I did just update all of my drivers yesterday and no change. BIOS is on latest version also.

 

 

Last but not least, the dip was no stutter, actually I hit the keys at that moment and didnt really focus on stutter in that specific splitt second, but I am pretty sure it was fluid as I had my son watching it and I told him what we did. So he would have said, "DAD, there is the stutter"..but he didnt.

 

This stutter is basically a frame drop. It happens in 1 second so if you blink you won't see it. If you're flying a normal mission you have to be focusing on something to detect it.

[bandit648]

So I did some more testing and ran the track with real-time monitoring like you did. See my attached screenshot, it looks exactly like yours. That first little dip after the start of the track is the stutter. If you enable logging and the OSD graph like I described in my previous post it will be easier for you to detect.

 

I'm 98% sure this is a DCS engine issue and not something I can solve. I've done similar tests in Prepar3d V3, AeroFly FS2, X-Plane 11, Battlefield 5 and Ace Combat 7 and none of them stutter with frametime spikes like DCS does.

[FragBum]

Are you hitting 100% on any single CPU core usage?

 

I did that test and got a solid 90FPS Rift VR.

 

As an observation if I carefully tweak settings to keep any single CPU core usage or GPU usage below 100% with some headroom DCS is buttery smooth. (I mostly fly rotary and low to the ground.)

 

Some map features seemingly runways* can affect FPS even from a goodly distance, for example in NTTR at the Hoover Dam I can easily get 90FPS until I bring the aircraft around to point at Las Vegas and FPS drops to 45, actually I can use this as a compass.

 

 

See TM screenshot from your test track note one core hitting 80% I'd be sure if that (or any other) core hit 100% FPS would drop.

 

Test done on 9600K(stock) with 2080Ti, 32GB 2800Mhz RAM. DCS overall CPU utilisation hit about 26% which is about average for this build.

[delevero]

I thought i add a little info.

 

If i remember correctly then wags spoke about VR and performance in a recent podcast.

Long story short, he spoke about that they want to optimize VR in dcs, and that they are aware that the terrain of dcs is a key area that really impact performance and that it need to be tweaked/updated to make dcs perform better.

 

Personaly i use the pimax 4k, and when parked in the f18c quick mission where we start in the airport ready to go, then I have lower fps than when i get up to example 5000ft.. I find that a bit odd since I would asume it would be more easy for the computer to show things from ground view in airport where nothing is "moving" while me flying a 500knots at 5000 feet and looking out the window.

 

Personaly though my pimax 4k and dcs have really good performance

( ryzen 2800x 8 cores 16 threats 16gb ddr 4 nvidia 2080rtx..

I had a 1080gtx before and it rand dcs good, but the 2080rtx have much more power and now dcs always run excellent in vr ).

[bandit648]

Are you hitting 100% on any single CPU core usage?

 

I did that test and got a solid 90FPS Rift VR.

 

As an observation if I carefully tweak settings to keep any single CPU core usage or GPU usage below 100% with some headroom DCS is buttery smooth. (I mostly fly rotary and low to the ground.)

 

Thanks for the suggestion. My CPU utilization is very low while running the test track. I get a spike to 28% on Core 12 but they generally all stay well below 20%. The frametime spike in this track occurs at 20 seconds after the start every time. It happens in just one second so its very easy to miss. You need to be logging perf data in MSI afterburner to see it.

 

I'm using this track because I found I could easily re-produce the issue with it. If I run the same scenario except just let the camera stay still, there's no frametime spikes. It's as if the frametime spike occurs as DCS is trying to read terrain object data from disk/memory.

[bandit648]

Little update on this... As a test, I installed DCS Stable branch on my Dell XPS laptop which has a GTX 1050 gpu. I had to put most settings on low to maintain a solid 60 fps but it runs fine at 1080p.

 

I ran the same free floating cloud test and sure enough.. ocasional frametime spike/fps drop while floating above the clouds. I was hoping this was something that could be fixed through a combination of settings and hardware but it looks like this is just a DCS engine issue.