?Unintended behaviour - VR MSAA Mask Size does not improve performance - 2.5.6

[Aarnoman]

Version: 2.5.6.49314, Standalone

 

Testing system used:

 

• Win 10 x64
  
• GTX 1080
  
• i7-9700k
  
• 16GB Ram
  
• DCS installed on SSD
  
• VR system: Oculus Rift CV1
  

 

 

 

 

Intended/Expected function: This version of DCS released the first iteration of MSAA mask size for VR. The stated function of this feature is to allow MSAA in the center of view, while leaving the peripheries free of AA. This serves the benefit of allowing for MSAA use in VR with a lower performance impact, as only the center of screen has to be antialised.

 

 

Bug/Problem: Currently, the performance is identical regardless of mask size (0.1 compared with 1.0). The mask does appear to work, as with mask set to 0.1 only the center of the screen is antialised.

However framerate remains identical with mask at 0.1 and 1.0 when the same mission is run. This leads me to conclude that the mask is applied after the MSAA pass. As a result, the mask does not serve a function, as there is no gain in performance with using it, but does reduce visual quality as the peripheries are not antialiased.

 

Test mission used: Fa-18c ready on ramp caucasus

 

 

Observed fps on this system: Average 34-38fps with MSAA turned on 4x. This remained at 34-38 fps with mask size set to 0.1, and mask size set to 1.0. The antialising was confirmed to be only in center of vision for 0.1, and full visual field when set to 1.0 - thus the mask is applied.

 

When MSAA is set to 0x, performance increased to 45fps (with vertical sync on).

 

 

Proposed solution: Mask should be applied prior to MSAA pass to allow for less resource use and thus performance improvement.

 

 

Conclusion: The current implementation of MSAA mask for VR serves no purpose, as the whole frame is AA before mask application, leading to identical resource use (and performance drop) regardless of what it is set to. However, with VR mask set to a low value, visual quality is degraded (as peripheral vision is not antialised).

 

 

 

 

Important note for replication - MSAA changes will not be applied until DCS is restarted.

Edited May 20, 2020 by Aarnoman

[FiringPin]

sometimes i wonder who tests these new features.

I also noticed the failure of this MSAA mask.

I hate to admit, but this sim is getting less and less desirable for VR users in particular

[sopebird]

I didn't see obvious difference (anti-alising and FPS) with different MSAA mask option.

HP Reverb, 150% super sampling in Steam VR, RTX 2080Ti.

[Aarnoman]

I didn't see obvious difference (anti-alising and FPS) with different MSAA mask option.

HP Reverb, 150% super sampling in Steam VR, RTX 2080Ti.

 

Did you restart DCS after changing mask settings? This is required for changes to apply. Give it a try without supersampling to better visualise the change.

[wju]

copy that - changing mask size is visible but has no effect to performance

[Hippo]

I definitely saw a difference in performance. See:

 

https://forums.eagle.ru/showpost.php?p=4341829&postcount=499

 

and my previous posts from yesterday in the thread.

 

 

 

 

 

 

EDIT: I thought I'd add the following. I tried Kegetys mod with a recent 2.5.6 version and noticed that I was not getting the improvement I was used to getting on previous versions (I used to get a 25% fps bump, now no more than 10%). I reported it on his thread, but it seemed to be just me, so I just let it go and forgot about it.

 

https://forums.eagle.ru/showpost.php?p=4296328&postcount=1323

 

Now I notice that with this DCS change, the improvement is also small. To OP: If you get the same sort of improvement I'm seeing, then you're probably going to get no more than about 3 or 4 fps, which you'll probably not be able to pick up in your measurements if you're measuring fps in the range 34-38. So maybe you were expecting a greater increase than is being delivered by this update?

 

This all makes me wonder if something has changed in the 2.5.6.x versions which makes the AA mask technique not as effective as it should be?

Edited May 22, 2020 by Hippo

[Sandman1330]

MSAA is very GPU intensive. Just a shot in the dark as I haven’t tried this myself yet, but is it possible your bottleneck is not the GPU, but maybe CPU, or RAM, or somewhere else?

 

If you have the GPU horsepower, the MSAA shouldn’t lower FPS much until the GPU is maxed out.

 

For OP, with a 9700K and 1080, it’s possible or maybe probable that you are GPU limited, so potentially there is an issue. Can you use Afterburner or another program to verify if your GPU is running near 100%?

Edited May 22, 2020 by Sandman1330

[Aarnoman]

Hippo, Sandman, please reference wju's graph. There is no difference regardless of mask size.

30fps drop with MSAA set to 4x (only change), and no difference in performance drop regardless of mask size. As I stated, mask size does not currently have a use.

Please provide evidence to the contrary comparing performance between mask size at 0.1 and 1 if you think there is any difference in performance.

Edited May 22, 2020 by Aarnoman

[Hippo]

Please provide evidence to the contrary comparing performance between mask size at 0.1 and 1 if you think there is any difference in performance.

 

 

If you look at the GPU usage % graph in my link, and in the corresponding ones in my previous posts for the same thread for mask 1.0, 0.42 and 0.1 you will see that I already done so:

 

 

1.0 - https://forums.eagle.ru/attachment.php?attachmentid=236376&d=1590017239

 

0.42 - https://forums.eagle.ru/attachment.php?attachmentid=236375&d=1590017239

 

0.1 - https://forums.eagle.ru/attachment.php?attachmentid=236465&d=1590059601

 

 

The graphs show me flying the same mission in the same way for a few minutes each time. The GPU usage % decreases (not by a lot), as the mask goes from 1.0 -> 0.42 -> 0.1. With 0.1 I can sustain 80 fps throughout the mission, with the the other two settings ASW (40 fps) has to kick in to varying degrees as performance cannot be sustained.

Edited May 22, 2020 by Hippo

[RoyMi6]

I can't see any difference even with just the mask size change, let alone performance.

 

 

Is anyone able to provide a screenshot that shows the mask size actually having any effect?

 

 

From HIPPO's own images I can't see a difference in the edges of the view where the mask should be "on" and "off" in theory.

 

 

[Hippo]

I can't see any difference even with just the mask size change, let alone performance.

 

Is anyone able to provide a screenshot that shows the mask size actually having any effect?

 

From HIPPO's own images I can't see a difference in the edges of the view where the mask should be "on" and "off" in theory.

 

 

 

 

(if you can be bothered) please see my other recent (post-patch) comments in this thread:

 

https://forums.eagle.ru/search.php?searchid=34570384

 

I have also replied to you in another thread.

[Aarnoman]

Curious, thanks for sharing hippo. I don't have a good explanation for why mine and wju's findings differ from yours; seems to work as intended and expected for you. Hopefully we will get more users data soon to figure out what is going on here.

[USA_Recon]

Curious, thanks for sharing hippo. I don't have a good explanation for why mine and wju's findings differ from yours; seems to work as intended and expected for you. Hopefully we will get more users data soon to figure out what is going on here.

 

There is a complete lack of documentation from ED on this as well.

 

1. Do you need MSAA enabled on page one (ie 2x)

2. It says 'bloom', some explanation would be good

3. I don't know if I see a difference or not, for me I can increase SS on SteamVR, however, any AA applied seems to do nothing (except use up 15-20fps hit)

[Hippo]

Hopefully we will get more users data soon to figure out what is going on here.

 

Let's hope so. I was hoping to get a bigger increase in performance from the AA mask setting. Turning off 2XAA completely gives me around a 33% increase, so I was hoping for around 20%. Something seems off.

[BMGZ06]

I am just sticking with steamvr SS set at 180%. Gives me a good enough res and sharpness and doesnt hit performance as hard as using in game MSAA even with this new mask. No clue what the the Bloom setting does. I unchecked and restarted and nothing looks different. The Ball in VR is still insanely bright until I get in close to the round down of the ship. before that the ball is so bright that I cant tell if it is centered or not. The new overlay is not something I like at all. It is distracting and I disabled it with an edit someone posted in the supercarrier section.

[Motomouse]

BGMZ06, what FPS do you achieve on the 2080Ti and the Reverb with your settings? Thank you.

[BMGZ06]

BGMZ06, what FPS do you achieve on the 2080Ti and the Reverb with your settings? Thank you.

 

If you are asking if I can a stable 45+ all the time the answer is no.

[Taz1004]

Bloom is basically glow. If you have shader mod, I believe it disables it.

To me, bloom is camera effect. Like motion blur or depth of field. You get these when you look through camera but usually not with naked eye.

[=Panther=]

Version: 2.5.6.49314, Standalone

 

Testing system used:

Win 10 x64

GTX 1080TI

i9-9900k

32GB Ram

DCS installed on M.2

VR system: Oculus Rift S (using steamVR SS=199% for pure testing purposes)

 

Prior to this patch I was receiving horrible FPS in a vfw mission: <15 FPS with several stutters and black displays in the HMD. It was the first time since going to VR that I lost all motivation to fly, it was just awful.

 

Patched to the latest version as indicated above and decided to try out the new adjustments. Using same configuration as prior to the patch with the MSAA off I saw FPS locked at 40FPS.

 

MSAA to 2x: 40FPS (Mask 1.0)

MSAA to 4x: 40FPS (Mask 1.0)

 

Adjusted mask value had no affect on performance of FPS but did see it actually working. A lowest value of .10 I could see a tiny clean box in front of my view. I adjusted the MSAA Mask while in the jet (it is live and adjustable from in the jet while in options menu). I slowly brought the value out to .45 which is where the jagged lines disappeared to the left and right of my rift s. Throughout the test FPS remained 40 both reported by DCS ingame counter, and the Steam fpsVR.

 

 

Here is example of my VR configuration, note this is only VR and graphics, not the entire config.lua.

["VR"] = {
	["bloom"] = false,
	["box_mouse_cursor"] = true,
	["custom_IPD"] = 58,
	["custom_IPD_enable"] = true,
	["enable"] = true,
	["hand_controllers"] = false,
	["interaction_with_grip_only"] = false,
	["msaaMaskSize"] = 0.45,
	["pixel_density"] = 1,
	["prefer_built_in_audio"] = false,
	["use_mouse"] = true,
},
["format"] = 1,
["graphics"] = {
	["DOF"] = 0,
	["LensEffects"] = 0,
	["MSAA"] = 2,
	["SSAA"] = 0,
	["SSAO"] = 0,
	["SSLR"] = 0,
	["anisotropy"] = 4,
	["aspect"] = 1.7777777777778,
	["chimneySmokeDensity"] = 1,
	["civTraffic"] = "",
	["clouds"] = 1,
	["clutterMaxDistance"] = 0,
	["cockpitGI"] = 0,
	["effects"] = 3,
	["flatTerrainShadows"] = 2,
	["forestDistanceFactor"] = 0.3,
	["fullScreen"] = false,
	["heatBlr"] = 0,
	["height"] = 1080,
	["lights"] = 2,
	["messagesFontScale"] = 1,
	["motionBlur"] = 0,
	["multiMonitorSetup"] = "1camera",
	["outputGamma"] = 1.2,
	["preloadRadius"] = 60000,
	["rainDroplets"] = true,
	["scaleGui"] = 1,
	["shadowTree"] = false,
	["shadows"] = 0,
	["sync"] = false,
	["terrainTextures"] = "max",
	["textures"] = 2,
	["treesVisibility"] = 6000,
	["useDeferredShading"] = 1,
	["visibRange"] = "High",
	["water"] = 2,
	["width"] = 1920,

[VitS]

VR system: Oculus Rift S (using steamVR SS=199% for pure testing purposes)

 

 

Wut?

[speed-of-heat]

I might be wrong but i think it is masking Anisotropic filtering?

[=Panther=]

Wut?

 

What is the confusion?

[VitS]

What is the confusion?

 

You running Rift S with SteamVR SS?

[=Panther=]

You running Rift S with SteamVR SS?

 

Yes, allows for better solutions when streaming to Twitch while in VR.

[VitS]

DCS uses Oculus apis, not Steam openvr. Are you saying that changing SteamVR SS somehow works with Oculus Rift S?