DCS Vr performance with 3090

[Lurker]

2 minutes ago, Avalanche110 said:

Having removed some pretty significant CPU bottle necks. I decided it would only be fair to do another comparison between the 1080ti and 3090.

 

Test Scenario:

Same mission, loaded onto a dedicated server running on the same PC I use for VR.

Flying a Gazelle at treetop level from a Fob in Karsi, through mountain valleys to a target near Allepo

I decided to be kinder to the 1080ti this time around and set the time of day to 8:30 as opposed to 7:30 (dawn) in the original test.

 

Test Methodology:

FPSVR Data logging utility to record  GPU & CPU frame times.

This method is far less taxing on the CPU than recording the flight in OBS.

 

Removing the CPU bottle neck really helped the 1080 achieve a smooth and consistent frame rate, where as it previously jumped around a lot more.

I was instantly reminded of why I decided I needed a new graphics card, average GPU frame times during cold start were in the mid 20's and It was barely making 40FPS.

The GPU utilization was pinned at 98% for the entire flight, It did manage to sneak into the 50fps range once I had reached the open planes behind Karsi.

The lowest GPU frame time recorded was 17.95ms. Lowest FPS recorded was 25FPS, highest was 53FPS.

AVERAGE GPU FRAME TIME = 22.62ms

AVERAGE CPU FRAME TIME = 10.20ms

AVERAGE FPS = 43.7

 

Swapped out the 1080ti and reinstalled the 3090.

Repeated the test using the exact same settings, drivers, resolution etc.

Was blown away at the improvement (again)  average GPU frame times approximately 10ms faster. Surprisingly, I did notice marginally better CPU frame times. 

This may have to do with the higher Vram on the 3090, or the drivers are running more efficiently with the newer Graphics card.  

There were several occasions when FPS was locked at 90 which is the Maximum Hz the Pimax can run at with a Large FOV setting. 

 

AVERAGE GPU FRAME TIME = 11.38ms (improvement:  98.76%)

AVERAGE CPU FRAME TIME = 9.88ms (improvement: 3.23%)

AVERAGE FPS = 76.0 (improvement: 73.91%)

 

I don't think I could possibly make this test more accurate or the evidence any more empirical.

if any one has any suggestions, please let me know.

 

 

 

 

 

 

Could you please elaborate on what settings specifically bottlenecked your CPU? As someone who also switched to a 3090 recently (up from the 2060) I'm still struggling to find the right settings for the best possible performance of the graphics card. 

[Avalanche110]

No problem.

1). When testing originally, I was using OBS to record the flight. that was quite taxing on the CPU.

2). Tacview, was a MASSIVE burden. as much as i like tacview, I uninstalled it.

3). Mirrors were using 0.6% of the CPU even though the gazelle does not have mirrors, disabled it in DCS settings.

4). Disabled "Fixed foveated rendering" in Pitool - this made a massive difference to CPU frame times and increased GPU load.

 

If you refer back to the original post in this thread, you can see the system specs and all the optimizations I have used.

All testing was done on a mission loaded to a dedicated server running on the same machine, this does wonders for smoothing out DCS in VR.

 

DCS SETTINGS.bmp

[Glide]

33 minutes ago, Avalanche110 said:

Fixed foveated rendering

Did you notice any visual difference?

[Avalanche110]

Just now, glide said:

Did you notice any visual difference?

The decrease in GPU frametimes make it significantly smoother.

I kept the settings the same in the test so it would be an accurate comparison between the GPU’s

[Glide]

6 minutes ago, Avalanche110 said:

The decrease in GPU frametimes make it significantly smoother.

I kept the settings the same in the test so it would be an accurate comparison between the GPU’s

Sorry, I meant, did you notice any less sharpness without foveated rendering?  My understanding is this was supposed to be an optimization by only sharpening the areas in front of your vision and not the peripherals.  Funny how it had the opposite affect, so I am wondering if there is any visual benefit to foveated rendering.

[Avalanche110]

On 3/11/2021 at 10:25 PM, Lurker said:

 

Could you please elaborate on what settings specifically bottlenecked your CPU? As someone who also switched to a 3090 recently (up from the 2060) I'm still struggling to find the right settings for the best possible performance of the graphics card. 

 

I will Direct you to SpeedOfHeats thread where he explains his optimizations and settings in Depth.

 

I happened to stumble across it yesterday and found it Really useful.

[Avalanche110]

On 3/11/2021 at 11:31 PM, glide said:

Sorry, I meant, did you notice any less sharpness without foveated rendering?  My understanding is this was supposed to be an optimization by only sharpening the areas in front of your vision and not the peripherals.  Funny how it had the opposite affect, so I am wondering if there is any visual benefit to foveated rendering.

 

Fixed Foveated Rendering is a way of reducing a VR applications GPU load.

It achieves this by rendering the outer portions of your FOV at a lower resolution.

I found that It created some really undesirable distortions and artefact's in the display. It also increased the CPU overhead, so their was no real advantage in using it at all.

I suspect that It might work better with Eye tracking. But in DCS for now. Its not desirable.

[frenzon]

FWIW (because it probably varies from PC to PC) I have the opposite experience with FFR on my Pimax 8KX / 3090 / 9900K - I love it - with aggressive FFR I can play Syria at native / 80% SS / Small FOV / VR defaults smoothly.

 

Without FFR the framerate drops into the red. For me, I'd rather have artifacts at the edge of my vision than doubled-up stuff in the center from motion smoothing.

 

Edited March 13, 2021 by frenzon

[Avalanche110]

3 hours ago, frenzon said:

FWIW (because it probably varies from PC to PC) I have the opposite experience with FFR on my Pimax 8KX / 3090 / 9900K - I love it - with aggressive FFR I can play Syria at native / 80% SS / Small FOV / VR defaults smoothly.

 

Without FFR the framerate drops into the red. For me, I'd rather have artifacts at the edge of my vision than doubled-up stuff in the center from motion smoothing.

 

 

Can I ask if your using the Vr Shader mod? if so what did you set the MaskSize to for the Pimax ( I have it set to 0.675)

I suspect that FFR does not play well with the rest of the optimizations I am using. It might be worth looking into.

Also,

How do you find the 8KX? I am considering upgrading from the 5k+ 

[frenzon]

5 hours ago, Avalanche110 said:

Can I ask if your using the Vr Shader mod?

 

I'm not - maybe that's a key difference. I had been using it in the past, but it wasn't clear 

 

Quote

How do you find the 8KX? I am considering upgrading from the 5k+ 

 

Ah, sorry I missed that you were on the 5K+; I had the 5K+ before the 8KX, and after much tweaking (e.g. Small FOV) to get to an OK framerate, could never go back - though the recent visibility changes might make the 5K+ actually better for spotting, with the 8KX I can spend a lot of time inside the cockpit happily dealing with MFDs (even the F-16's tiny MFDs). If you don't care as much about tiny text, it might not be as valuable an upgrade.

Small FOV sounds like a waste of the Pimax's strengths, but it's still huge compared with the G2 and others.

 

[Go2Matt]

So FPS aside... Coming from an1080ti can you turn the eye candy up considerably. 

I've recently upgraded from a rift to a reverb G2. Not really played since the upgrade as I've been on the lookout for either a 3080 or a 3090. If I've not got a 3090 by the time the 3080ti comes out I may get one of those..

Or glasses

 

Edited March 13, 2021 by Go2Matt

[Avalanche110]

6 hours ago, frenzon said:

 

I'm not - maybe that's a key difference. I had been using it in the past, but it wasn't clear 

I have DCS:Stable installed without the shader mod. I tried FFR on the Pimax and didn't notice any undesirable distortions. 

It would be interesting to see what delivers the biggest performance increase. FFR or the shader mod.

[Avalanche110]

2 hours ago, Go2Matt said:

So FPS aside... Coming from an1080ti can you turn the eye candy up considerably. 
 

Yes!....... and No!

There are so many DCS settings that provide little to no visual improvement, but are a massive strain on your system.

I ran another test in the Gazelle to see how far I could push the settings while maintaining a smooth 45FPS.

 

PD set to 1.0 

SteamVR resolution to 200% 

 

I noticed only a slight improvement in visual quality over my original settings. 

Terrain textures set to high uses 20Gb of VRAM, thats double what a 3080 has, Provides 0 improvement in visual quality ( in my opinion )

MSAA at 4X is slightly better, but not worth the FPS hit.

SSAA and SSLR, didnt notice any difference with them on, other than massive GPU load.

Anisotropic filtering at 16X: you have to look pretty hard to notice a difference.

Visible Range is the real Killer! I use Medium for helicopters, high for fixed wing. 

GPU utilization was very inconsistent.  96% in forested areas, and 64% in open areas.

I will definitely be sticking with my original settings.

 

When I was on the 1080ti, I would prioritise Good visual quality at angels 30 and put up with lag and stutter when taxiing to the runway.

Upgrading to the 3090 provides the eye candy at altitude, and has dramatically improved the experience on the ground.

 

[dburne]

3 hours ago, Go2Matt said:

So FPS aside... Coming from an1080ti can you turn the eye candy up considerably. 

I've recently upgraded from a rift to a reverb G2. Not really played since the upgrade as I've been on the lookout for either a 3080 or a 3090. If I've not got a 3090 by the time the 3080ti comes out I may get one of those..

Or glasses

 

 

 

So a lot depends on your overall rig as well and not just the GPU alone.

I run an EVGA RTX 3090 FTW3 Ultra, but I also have an i9 9900k@ 5.2 GHz on all cores feeding it along with 32 Gb of CL14 3200 MHz ram.

I run DCS with most graphics on highest at 100% Steam VR resolution for Reverb G2. No MSAA as I don't need it at this resolution.

I have Motion Smoothing on and forced at 45 fps. Extremely smooth and stutter free for me, playing single player campaigns with the Hornet.

 

I am very happy with both the looks and the performance in DCS with my Reverb G2.

[Avalanche110]

33 minutes ago, dburne said:

 

So a lot depends on your overall rig as well and not just the GPU alone.

 

Can I ask what GPU you upgraded from? Did you notice a significant performance increase going to a 3090.

I have found the 3090 to be a night and day when compared to the 1080ti

[dburne]

5 minutes ago, Avalanche110 said:

Can I ask what GPU you upgraded from? Did you notice a significant performance increase going to a 3090.

I have found the 3090 to be a night and day when compared to the 1080ti

 

Yes I upgraded from a 2080 Ti FTW3 Ultra.

However hard to say on the performance department as I upgraded from Rift S to a Reverb G2 at about the same time.

[GunSlingerAUS]

On 3/11/2021 at 11:08 PM, Avalanche110 said:

 

 

2). Tacview, was a MASSIVE burden. as much as i like tacview, I uninstalled it.

 

 

DCS SETTINGS.bmp 4.37 MB · 15 downloads

 

Interesting - do you have any numbers to back this up? I've read that it's at most a 3% CPU overhead on a decent CPU these days - so to hear it's a MASSIVE burden goes against everything the developer and all supporting documentation claims. I use it during our Squad Op days where we have over 30 human aircraft, plus another 500 or so AI entities - nobody in the squadron has noticed any issues?

[Avalanche110]

4 minutes ago, GunSlingerAUS said:

 

Interesting - do you have any numbers to back this up? I've read that it's at most a 3% CPU overhead on a decent CPU these days - so to hear it's a MASSIVE burden goes against everything the developer and all supporting documentation claims. I use it during our Squad Op days where we have over 30 human aircraft, plus another 500 or so AI entities - nobody in the squadron has noticed any issues?

 

In most circumstances, Tacview will use very little CPU overhead.

I have used it for years without issue. However.....

Tacview's CPU usage Scales with the size of the mission. the more units and munitions that tacview needs to track, the more CPU overhead it will consume.

I Run a dedicated server with a mission running hundreds of lua scripts, and hundreds of units. The Server log indicated that :GetWorldObject

( just 1 algorithm Tacview requires)  was using 7% of CPU. this was running on both the Server and the OB install so tacview's impact was significant.

Uninstalling Tacview improved CPU frame times by a noticeable margin. I'm not suggesting that anyone else immediately uninstall it, Just be aware that it can have a significant performance impact in some circumstances.

 

As a reference the average Tacview file that i would review after running the server would be between 15Mb & 18Mb

[Avalanche110]

15 hours ago, frenzon said:

 

I'm not - maybe that's a key difference. I had been using it in the past, but it wasn't clear 

 

I did a little more testing on FFR.

I got FFR working with the shader mod without significant artefacts. Unfortunately it increased CPU frame times by around 5ms on average and decreased frame rates.

GPU utilization was significantly lower with FFR, this would definitely be an advantage for a lot of people.

 

For me however, I have opted to stick with FFR off.

[Shrike_AU]

I did quite a bit of testing with my 5k+ and the eye tracking / Foveated rendering in DCS when I first got the eye tracking working. On my rig at the time (which was heavily CPU bound) it made no difference. You can test this without the eye tracking by running a benchmark mission (the Su25 benchmark mission in User Files on this website is excellent for this) with the headset stationary. The FPS between FFR on and Off were negligable.

 

@Avalanche110 I'd be interested to see what improvement you get with Terrain Object Shadows at flat. That makes a big difference for me.

 

In any case, I'll have a look at lower PD and higher Steam SS and see what I get.

 

Thx for the analysis.

 

[update]

 

I spent the morning playing around and ended up changing from PD of 1.0 and steamvr per app SS of something like 60% to PD 0.7, SteamVR Res 100%, Per App SS 300%. Running the headset at 110Hz, gives around 60+ FPS in most tested scenarios, so with motion smoothing on to lock FPS at 55, it's pretty nice.

 

Edited March 17, 2021 by Shrike_AU
Update

[Avalanche110]

9 hours ago, Shrike_AU said:

I did quite a bit of testing with my 5k+ and the eye tracking / Foveated rendering in DCS when I first got the eye tracking working. On my rig at the time (which was heavily CPU bound) it made no difference. You can test this without the eye tracking by running a benchmark mission (the Su25 benchmark mission in User Files on this website is excellent for this) with the headset stationary. The FPS between FFR on and Off were negligable.

 

@Avalanche110 I'd be interested to see what improvement you get with Terrain Object Shadows at flat. That makes a big difference for me.

 

I imagine FFR would be very useful in a situation where the GPU was the limiting factor, AND their was CPU overhead. 

But From what I could tell, It made it look and run worse. @frenzon reported that it made a huge improvement running his 8kx.

This technology is not supported by older GPU's. The FFR options were not available in Pitool when the 1080ti was installed.

 

I had terrain shadows set to flat for along time. It wasn't until I changed P.D settings that went to Default shadows.

I recently started using this Shadow mod I found in SpeedOfHeats Thread. give it ago and see what you think.

 

 

[Avalanche110]

I have just pre-ordered an 11900K, along with a all the bits for a new build.

I will do another set of comparison tests to objectively quantify what improvement it makes.

If anyone has suggestions on a test scenario, or a particular mission, or a particular module, please let me know.

 

That's if I actually get it this year.....

[speed-of-heat]

18 minutes ago, Avalanche110 said:

I have just pre-ordered an 11900K, along with a all the bits for a new build.

I will do another set of comparison tests to objectively quantify what improvement it makes.

If anyone has suggestions on a test scenario, or a particular mission, or a particular module, please let me know.

 

That's if I actually get it this year.....

it really doesnt matter.. providing you do something consistent 

[nobuttons]

16 hours ago, Shrike_AU said:

I did quite a bit of testing with my 5k+ and the eye tracking / Foveated rendering in DCS when I first got the eye tracking working. On my rig at the time (which was heavily CPU bound) it made no difference. You can test this without the eye tracking by running a benchmark mission (the Su25 benchmark mission in User Files on this website is excellent for this) with the headset stationary. The FPS between FFR on and Off were negligable.

@Avalanche110 I'd be interested to see what improvement you get with Terrain Object Shadows at flat. That makes a big difference for me.

 

I have a Pimax 8K X, and Fixed Foveated Rendering makes a *huge* difference for me. I'll do some benchmarks on it, but it's like +10fps! I'd run it on every game if it didn't produce glitchy rendering artifacts on half of them. I did order an eye tracker to see if it is better, but haven't gotten to try it yet. What I see in FPS VR is that it dramatically lowers GPU load, so yeah not surprising if you are CPU bound that it has no impact.

 

On my rig many of the other recommended tricks have had zero impact on my bechmarks. VR shaders, multithreading, Windows tweaks, non of them produced *any* measurable impact.

 

On the flip side the most impactful DCS settings are shadow quality and MSAA quality for sure! By tuning thooe I feel like I have pretty darn good visual quality while keeping a steady "reprojected or better 75". 

DCS: Textures & Terrain High, Water & Vis Medium, Heat Blur Low, Shadows Medium, MSAA 2x, DoF OFF, Lens Flare Only, Motion Blur Off, SSAA & SSLR Off, Clutter 170, Trees 66%, Smoke 0, Anisotropic 2x, Terrain Shadows Flat, Cockpit Global Illumination On

 

To be fair I don't have the patience to run a 3 min benchmark between configuration tweaks, so I just use F16, Syria, Cold and Dark Ramp Start as my stable place to grab a 30 second sample for fps VR (which gives me 55fps when I also am running Discord, and Simshaker, and Leapmotion). Also that pesky desire to like actually fly missions keeps getting in the way...

 

I still gotta try the various mixes of Steam, PiTool and DCS resolutions since I've been putting off fussing with that variable

[Avalanche110]

16 hours ago, nobuttons said:

 

Also that pesky desire to like actually fly missions keeps getting in the way...

 

 Yes, I am quite familiar with that. 

 

16 hours ago, nobuttons said:

 

I still gotta try the various mixes of Steam, PiTool and DCS resolutions since I've been putting off fussing with that variable

I found using Pitool S.S was extremely taxing. There was noticeable a improvement in stuttering when rendering the same number of pixels using SteamVR.

I have attached a P.D Calculator so you can work out how many pixels your actually rendering. 

PDSSCalc.zip