My 3090 Settings for my G2 (2.7.11)

[dureiken]

Hi pilots

made some new tests on this track : https://www.mediafire.com/file/7oujviutzh6qqba/DCS_F16_Perf_test_Syria_SA6.trk/file

I use TGP and maverick which is really GPU time consumming

My results :

with these settings :

We can see that FSR is really GPU friendly but in VR for F16 I find that it really "blurries" (don't know the term sorry) the fonts. I will try to stay without FSR

 

Hopes it can help

[X_legio]

FSR ?

[speed-of-heat]

6 hours ago, X_legio said:

FSR ?

It’s also in the original post

[DroptheHammer]

@speed-of-heat

In the current OP version you mention "shader cache to 10GB" twice, but I cant seem to find how or what you did for that.  Could you elaborate?

[speed-of-heat]

it's an NVCP Global setting

Edited November 29, 2021 by speed-of-heat

[metalnwood]

I am sure that while reading about the mods etc that I read the potential of an issue I now have, except now I have the issue I can't find it

I was having some performance issues so I used skatezillas app to bring dcs openbeta to default.  I deleted files in the metashaders and fxo.

I installed the dcs shaders mod for vr with the version 2.7.7.14727   and added the shadows reduced fram rate impact.zip.

Ran DCS and everything was smooth again, good times

Then whenever I went to something that was dark, the outside was completely weird, bright and the ka50 was missing a lot of its textures so the cockpit was just a bright untextured model.

I am sure I read about this but would appreciate if someone could remind me of the cause.

 

Thanks!

[speed-of-heat]

"completely weird" covers quite some scope... but if its what i think it is; running the mission without the shaders and then  try again with the shaders it seemed to clear up some strange night effects

[metalnwood]

Thanks speed   Without being near the computer to take a screenshot I was a bit lost for words to describe it!  

    Great, I will remove and run a mission, if cleared up hopefully it will remain that way if I reapply them.

 

Thx.

[speed-of-heat]

seemed to work that way for me, when i had a similar experience on a night shot from a carrier

[ISantus]

I believe this is my first post, but I have been stalking these forums for about a year. I wanted to say THANK YOU to SOH, and others, for this very detailed and coherent process for tuning DCS. I have been working with my system with a Valve Index, but recently wanted to switch to a HP Reverb G2 with the recent updates to HP's hardware. As a result I found myself on this page tuning my system. It has been incredibly helpful in understanding my performance and making decisions to achieve what I wanted. Ironically my settings ended up almost exactly following SOH's recommendations. However, I wanted to share my findings for others that perhaps find their way to this page. 

System specs: 5900x (no overclock yet) on Asus x570 Crosshair Hero VIII, 3090 FE with slight overclock, 32gb at 3600, WIndows 11, MWR Steam VR (non-beta), HP Reverb G2. Running latest Nvidia drivers and DCS Beta fully up to date. 

I tested the difference between the multiple mods he mentions, as compared to none, as well as my final decision of VR Shaders and FSR. I also wanted to test if vector reprojection in WMR vs Steam reprojection made any difference. It did not and I settled on Steam VR with WMR defaulting to Steam. From there I tested between MSAA at x2 vs off, terrain shadows flat vs off, standard clouds vs ultra clouds, as well as cockpit global illimunation on vs off. For sake of argument I also wanted to see if my GPU overclock helped. I ended up cheating and using MSI Afterburner OC Scan to settle on a core Curve and +300 on the memory. It did make a slight difference. I think I can push the card (can you really call GPUs these days cards?) further, but that is for another time in addition to overclocking my CPU and continuing my tests. 

I completed this entire process over about 70 tests and 2 different scenes. I tend to fly off the SC, but I realize it is overlly demanding as compared to other scenes. To test the SC I used the Mission SC cold start in the Persian Gulf. I used this as kind of a worse case scenario. I also fly in the Persian Gulf a lot so I used an instant action free flight scene to test a more traditional situation. To get fair results I used a very exact moment in time to note the current frame timing in each of these scenes. For the SC I used the moment the first F18 takes off from the deck and marked my timings. For the Persian Gulf free flight, the second I can see the white buildings on the right side of the lake pass the cockpit, I marked the timings as my score. To be fair, my measurements were not precise and since the frame timings move so much, I chose not to mark the decimal places. I simply chose the whole number as the timing. Yes, this isnt super precise, but it got me what I was looking for. Also of note, all of this was done from the cockpit of the incredible F18. 

Two screenshots attached for reference: System Settings as well as my spreadsheet for tracking the timings across these tests. In addition, I am running VR settings of PD 1.0 and mask of .42. I kept PD at 1.0 and adjusted between 80% and 100% via Steam settings for the tests as noted in my spreadsheet. Please note that the settings screenshot I have attached shows all of my other settings, but you can use my spreadsheet to determine changes I tested, for the remainder. Everything else besides what is noted on the spreadsheet were left as you see them in this screenshot. 

I wouldn't say this process was without some bordem, but it really felt good to undertand the impacts of each settings on frame timings. This allowed me to settle on my final settings: 80% SS, MSAAx2, Ultra clouds, low shadows, terrain object shadows off, cockpit global illumination off. All other settings noted in screenshot. 

Thanks again for all the amazing work to help me get a grasp on these settings. I hope others find my spreadsheet helpful. Now off to more tests and to overclock this CPU. Also, keep Oxford, MI on your minds and hearts. I live here 6 miles from the shooter's house. This community is shook. 

 

Edited December 4, 2021 by ISantus

[speed-of-heat]

@ISantus, I’m really glad the approach is working for you. Stay safe!

And thank you for such war and detailed feedback

[Xerno]

10 hours ago, ISantus said:

I believe this is my first post, but I have been stalking these forums for about a year. I wanted to say THANK YOU to SOH, and others, for this very detailed and coherent process for tuning DCS. I have been working with my system with a Valve Index, but recently wanted to switch to a HP Reverb G2 with the recent updates to HP's hardware. As a result I found myself on this page tuning my system. It has been incredibly helpful in understanding my performance and making decisions to achieve what I wanted. Ironically my settings ended up almost exactly following SOH's recommendations. However, I wanted to share my findings for others that perhaps find their way to this page. 

System specs: 5900x (no overclock yet) on Asus x570 Crosshair Hero VIII, 3090 FE with slight overclock, 32gb at 3600, WIndows 11, MWR Steam VR (non-beta), HP Reverb G2. Running latest Nvidia drivers and DCS Beta fully up to date. 

I tested the difference between the multiple mods he mentions, as compared to none, as well as my final decision of VR Shaders and FSR. I also wanted to test if vector reprojection in WMR vs Steam reprojection made any difference. It did not and I settled on Steam VR with WMR defaulting to Steam. From there I tested between MSAA at x2 vs off, terrain shadows flat vs off, standard clouds vs ultra clouds, as well as cockpit global illimunation on vs off. For sake of argument I also wanted to see if my GPU overclock helped. I ended up cheating and using MSI Afterburner OC Scan to settle on a core Curve and +300 on the memory. It did make a slight difference. I think I can push the card (can you really call GPUs these days cards?) further, but that is for another time in addition to overclocking my CPU and continuing my tests. 

I completed this entire process over about 70 tests and 2 different scenes. I tend to fly off the SC, but I realize it is overlly demanding as compared to other scenes. To test the SC I used the Mission SC cold start in the Persian Gulf. I used this as kind of a worse case scenario. I also fly in the Persian Gulf a lot so I used an instant action free flight scene to test a more traditional situation. To get fair results I used a very exact moment in time to note the current frame timing in each of these scenes. For the SC I used the moment the first F18 takes off from the deck and marked my timings. For the Persian Gulf free flight, the second I can see the white buildings on the right side of the lake pass the cockpit, I marked the timings as my score. To be fair, my measurements were not precise and since the frame timings move so much, I chose not to mark the decimal places. I simply chose the whole number as the timing. Yes, this isnt super precise, but it got me what I was looking for. Also of note, all of this was done from the cockpit of the incredible F18. 

Two screenshots attached for reference: System Settings as well as my spreadsheet for tracking the timings across these tests. In addition, I am running VR settings of PD 1.0 and mask of .42. I kept PD at 1.0 and adjusted between 80% and 100% via Steam settings for the tests as noted in my spreadsheet. Please note that the settings screenshot I have attached shows all of my other settings, but you can use my spreadsheet to determine changes I tested, for the remainder. Everything else besides what is noted on the spreadsheet were left as you see them in this screenshot. 

I wouldn't say this process was without some bordem, but it really felt good to undertand the impacts of each settings on frame timings. This allowed me to settle on my final settings: 80% SS, MSAAx2, Ultra clouds, low shadows, terrain object shadows off, cockpit global illumination off. All other settings noted in screenshot. 

Thanks again for all the amazing work to help me get a grasp on these settings. I hope others find my spreadsheet helpful. Now off to more tests and to overclock this CPU. Also, keep Oxford, MI on your minds and hearts. I live here 6 miles from the shooter's house. This community is shook. 

 

 

Are you using FSR as well with these settings?

[Csgo GE oh yeah]

Wow what an amazing labour intensive write up. 

In all honesty though i skimmed through it, i saw 80% in steam VR resolution. Oof i tried 125% today and quickly switched back to 150% though. 
How do you manage with 80% ? 

I guess i'll have to read the rest of it now  

[Steel Jaw]

very nice work, thanks

[speed-of-heat]

@Xerno

  

12 hours ago, ISantus said:

 

I tested the difference between the multiple mods he mentions, as compared to none, as well as my final decision of VR Shaders and FSR.

 

@Csgo GE oh yeah I typically run at 100% with FSR today, but i still test at 80% for a numbers baseline point of view, i still find myself using 80% in very intensive maps ... i honestly don't notice the difference after a few minutes

[ISantus]

48 minutes ago, Csgo GE oh yeah said:

Wow what an amazing labour intensive write up. 

In all honesty though i skimmed through it, i saw 80% in steam VR resolution. Oof i tried 125% today and quickly switched back to 150% though. 
How do you manage with 80% ? 

I guess i'll have to read the rest of it now  

To be honest, at this point my system can't cut it so I don't want to know what anything more than 100% looks like. I will be spoiled and then cut elsewhere. I very much enjoy MSAA and ultra clouds and as a result there is no overhead room in the SC (my preferred launch location). I am not CPU bound, but rather GPU bound, which is crazy considering it is a 3090. I either need to wait for software upgrades, or chase another GPU with the 4000 series around the corner. I am not counting on either. 

[Csgo GE oh yeah]

Clouds on Ultra in VR ? 
Why would you do that, they look horrible either way don't they ? 

[speed-of-heat]

they look considerably worse if not on Ultra, once you drop down to high its pretty much of muchness between high and medium 

 

[Csgo GE oh yeah]

In VR? Does it actually make any difference at all ? I'm skeptical i go test now

Well you're right there is a difference but they still look horrible in VR. The difference is certainly not enough to justify losing 15% performance , good lord. And that was at "light scattered". 

No for me the clouds stay on low, and to OFF whenever i can. 

Edited December 4, 2021 by Csgo GE oh yeah

[Ready]

Nice sheet! Let me also share mine, may it be of help to someone. 

All generic system optimizations as reported by the gurus (like SOH, Thud, Lukas S. and others) were done prior to starting with the testing. These were optimizations like W10 updating, pagefile optimization, SSD configuration, DCS proces priority configuration, Windows services optimization, HPET, gamebar, hotplug, dynamic tics, XMP profile and hyperthreading.

All changes made within DCS were tracked, also within NVIDIA control panel, WMR, various DCS mods, GPU and CPU overclocking. For OB updates and driver updates I kept track of the versions.

So far ran 55 tests. The first 26 tests used SOH's track file no. 276, after that used my own track file which I created in the hog to stay as close to my situation as possible. I usually run a test after making 1 modification to see the impact and sometimes go back and forth (switch back to the prior setting and retest, redo the setting and test again) just to confirm.

The changed setting is every time marked in red. The actual impact is displayed in lines 5-8, using earlier discussed method in FPSVR to capture in CSV file and calculating performance averages across the whole track file, which to me give a very fair impression. Sometimes also note down the reprojection ratio as observed in FPSVR on screen, but this is not very stable I must say. I have no idea why that is.

Also created a chart to visualize the DCS performance on my rig over time. GPU framerate (the orange line) is steadily improving it seems, even though shader mods become less effective because of game resource encryption. The graphics look good enough for me with the current settings. I judge them by my personal experience within the pit (inside VR) and by looking at the recordings on Youtube to see if it is a smooth experience (Youtube does quite some compresssion and downsampling).

The next couple of tests will be focused on cloud settings. (Current clouds are better than those circling clouds of older DCS versions but to me the current distant clouds are really ugly and a big distractor, so I think to turn them all the way down to gain a bit more performance. If only there was a way to stop the trees from growing and moving while flying over them.. This would make me a happy simmer.)

Another thing I don't fully understand is my GPU is never really utilized >90%, as reported in FPSVR. So there should be more performance to be gained by utilizing the final 5-10%? I just don't know how to do this. The performance is bound by GPU. GPU is in orange, CPU is usually in the green. 

See my rig specs in my signature.

Edited December 5, 2021 by Ready

[Steel Jaw]

HOW EXACTLY to do this please.

 

To ensure your DCS.exe, vrserver.exe and vrcompositor.exe are working most efficiently with your CPU set their priority to High, you can do this manually, or via a script I do it via the registry, add the following keys to the registry:

 

HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows NT\CurrentVersion\Image File Execution Options\DCS.exe\PerfOptions

"CpuPriorityClass"=dword:00000003

Edited December 6, 2021 by Steel Jaw

[tomeye]

16 hours ago, Ready said:

Nice sheet! Let me also share mine, may it be of help to someone. 

All generic system optimizations as reported by the gurus (like SOH, Thud, Lukas S. and others) were done prior to starting with the testing. These were optimizations like W10 updating, pagefile optimization, SSD configuration, DCS proces priority configuration, Windows services optimization, HPET, gamebar, hotplug, dynamic tics, XMP profile and hyperthreading.

All changes made within DCS were tracked, also within NVIDIA control panel, WMR, various DCS mods, GPU and CPU overclocking. For OB updates and driver updates I kept track of the versions.

So far ran 55 tests. The first 26 tests used SOH's track file no. 276, after that used my own track file which I created in the hog to stay as close to my situation as possible. I usually run a test after making 1 modification to see the impact and sometimes go back and forth (switch back to the prior setting and retest, redo the setting and test again) just to confirm.

The changed setting is every time marked in red. The actual impact is displayed in lines 5-8, using earlier discussed method in FPSVR to capture in CSV file and calculating performance averages across the whole track file, which to me give a very fair impression. Sometimes also note down the reprojection ratio as observed in FPSVR on screen, but this is not very stable I must say. I have no idea why that is.

Also created a chart to visualize the DCS performance on my rig over time. GPU framerate (the orange line) is steadily improving it seems, even though shader mods become less effective because of game resource encryption. The graphics look good enough for me with the current settings. I judge them by my personal experience within the pit (inside VR) and by looking at the recordings on Youtube to see if it is a smooth experience (Youtube does quite some compresssion and downsampling).

The next couple of tests will be focused on cloud settings. (Current clouds are better than those circling clouds of older DCS versions but to me the current distant clouds are really ugly and a big distractor, so I think to turn them all the way down to gain a bit more performance. If only there was a way to stop the trees from growing and moving while flying over them.. This would make me a happy simmer.)

Another thing I don't fully understand is my GPU is never really utilized >90%, as reported in FPSVR. So there should be more performance to be gained by utilizing the final 5-10%? I just don't know how to do this. The performance is bound by GPU. GPU is in orange, CPU is usually in the green. 

See my rig specs in my signature.

 

I have been thinking of this too, anyone knows if the percentage reported by fpsvr is for DCS or global? If it is the usage by DCS it would mean that rest 5-10% is used by OS and other processes.

[speed-of-heat]

2 hours ago, Steel Jaw said:

HOW EXACTLY to do this please.

 

To ensure your DCS.exe, vrserver.exe and vrcompositor.exe are working most efficiently with your CPU set their priority to High, you can do this manually, or via a script I do it via the registry, add the following keys to the registry:

 

HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows NT\CurrentVersion\Image File Execution Options\DCS.exe\PerfOptions

"CpuPriorityClass"=dword:00000003

 

I manually add them to the registry using regedit

 

[tomeye]

not sure if this helps, but I have the reg exported with steam vr. You can view with notepad (right-click) or apply to your system with double-click.

 

DCS High CPU.reg

Edited December 6, 2021 by tomeye

[Steel Jaw]

11 minutes ago, speed-of-heat said:

I manually add them to the registry using regedit

 

Right but pretend Im a complete moron and gimme a step by step.  I can add an actual reg key file but this requires CREATING the entry.

Edited December 6, 2021 by Steel Jaw