Pixel density vs MSAA vs supersampling: mid-level GPU

System: Ryzen 7, 16Gb RAM, Vega 56 on stock settings (should be about equivalent to a GTX1070)

Question: does the Oculus tool, and the ability to set supersampling, accomplish anything?

Findings:

1. I want a 1080ti

2. Oculus tool vs in-game settings is pretty much a wash

Endpoint: Today I'm mainly looking at the cockpit. How easy is it to read numbers, gauges and labels

For this test, I used the Oculus tool HUD and monitored App frame rate and App GPU render time. What I'm shooting for is a solid 45fps, no stuttering and no puking. Oculus wants to lock the frame rate in at 45; it seems the GPU render time is a more granular measure. For me, keeping that number less than or equal to 20 is the key to a smooth flight.

I dunno what "0.5" or "1.3" means when it comes to supersampling. I think it depends on what the developer wants it to mean. From what I've been able to surmise, the Oculus needs just a touch of supersampling at baseline. So fractions like "0.5" or "0.8" *probably* mean the app is rendering native resolution. "1" means the default level of supersampling, like you see in your little Oculus house. And anything greater than 1 is some multiple of default.

Pixel density in DCS and supersampling in the Oculus interact. Probably because it's the same thing. Near as I can tell, they multiply. So for example, if you set PD at 1.2 and SS at 1.5, your GPU bursts into flames. Just kidding! But your jet will, as it lawndarts into the ground while you're staring at the hourglass of death. I'd do one or the other meself.

Other settings:

ASW auto

Textures high

Terrain texture low (kinda looks the same thru Oculus)

Water high (why not)

Shadows off (Still undecided on this one)

Vis range medium

Anisotropic filtering off. Can't see that it does anything.

Setting: I'm flying the A4 in "free flight" over Caucasus. So, no other units on map. Procedure is to check performance up high, dive down over the city, and fly across town at rooftop level, and see if the fps holds up.

Data:

MSAA PD Supersampling FPS GPU Comment

off 0.5 0 too terrible for words

2x 1.0 0 up high: 45 15 gauges are readable but blurry

down low: 45 19 nice and smooth

2x 1.2 0 high: 45 19 looks about the same

low: 35 22 terrible stuttering

4x 1.0 0 high: 45 20 looks gorgeous

low: 35 29 severe stuttering, uncomfortable

4x 0.5 1 high: 45 19 looks OK, not as good

low: 42 22 some stuttering, not too bad

off 0.5 1.5 high: 45 18 looks almost as good as 4x/1.0/0

low: 42 19 some stuttering

off 0.5 1.3 high: 45 15 acceptable graphics

low: 45 19 smooth as silk

off 1.3 0 high: 45 15 see comment below

low: 45 20 smooth as silk

Tell ya this game looks awesome with MSAAx4. Instruments seem sharp, colors seem to pop. But adding in "stock" supersampling -- and Oculus needs a little -- just flogs the daylights out of my GPU. I could live with it flying around up high all by myself. But what's the fun of that?

On the other hand, if you take away MSAA, it doesn't take much supersampling to make up some of the difference, and for whatever reason that seems to go down better than the combo. Not as pretty but functional. Between working the in-game settings and letting the oculus tool do it, I think the oculus gives me a slightly sharper picture for the same cost. In other words I get the feeling it's just a tad more efficient. But it's darn close.

If I can figure out how to take screen shots in Oculus, I'll post up a comparison of those last two settings for ya. And I guess 2x/1.0/0 which is also a workable combo.

So bottom line is, MSAA 2x with default PD, or running MSAA off with PD or SS at 1.3 all seems to run about the same, and all seems to look about the same too. Not sure the Oculus tool adds much, maybe a little. I will say that you can make finer adjustments with either PD or Oculus tool than you can with MSAA, so it should be easier to find the sweet spot. For example to make room for those darn shadows. Or other units even.

At any rate. If you're breaking in a mid-level card, I'd try one of those combos to start with.

Interesting finding. thanks for taking the time. I'll give the 4X a whirl

Pixel density vs MSAA vs supersampling: mid-level GPU

very interesting!

thanks for taking the time to test this.

I'll add my thanks for this post as well . With a Rift on the near horizon , i find this the most useful post in the entire VR forum - and i have devoured every one !

The shader mod gives a bit of a boost to the GPU for increased PD. https://forums.eagle.ru/showthread.php?t=215373

I run 1.4 PD on high settings (bar shadows-flat) with 4x MSAA on a OCèd RTX 2080 & i7 4790k 4.6ghz, 24 gigs ram. The main issue is cpu optimization is poor and chokes the GPU when the map is busy. Again anything from 45-90fps. Usually 90fps when at altitude 1000ft+. Treetops 45-50fps. Nevada map can hover around the 90fps close to the ground.

Yes I have found the same. Mixing PD with MSSA at any level is not a good idea. 2X MSSA, 1.0 PD, is about as good as balance as I can find peformance wise. 4x MSSA looks nicer obviously but not worth performance hit with larger engagements. Still can't realistically fly epsom campain or anything on that scale, but can actually have a smooth experience when there are alot of planes around.

With MSSA PD off, it runs great but looks bad, really wish I could ignore it.

What about Anisotropic filtering? (the one on the right side of settings menu) What do you guys set that at?

Looks like this thread has been hijacked by not-mid-level gpus :)

Looks like this thread has been hijacked by not-mid-level gpus :)

Unfortunately with VR we are all mid-level GPU's!

What about Anisotropic filtering? (the one on the right side of settings menu) What do you guys set that at?

I think pretty high (8-16) as it doesn't have much of an impact. Doesn't make much of a difference sadly either!

I think pretty high (8-16) as it doesn't have much of an impact. Doesn't make much of a difference sadly either!

Thanks. I wasn't sure if it did anything.

Something to also use with your testing is task manager and look at the different loadings that the various settings incur.

I found I preferred MSAA to PD even if that meant dropping other settings to med/low etc.

I found PD because it artificially increases the resolution and the artefacts it creates increased the amount of work the MSAA filter had to do, which makes sense but that amount of work increases in a nonlinear fashion. This is just by observing what the effects of MSAA and PD were on GPU and CPU usage.

You might be able to improve the VR image by using MSAA at 4 and PD at 1.1 even if that means lowering some others. I also find to keep the VR frame rate at 45 with ASW was also to not push CPU and GPU at 99%+ but rather CPU at around 80% average and GPU at around 70% (that kinda works for me you may need to adapt for your system) However doing that seems to give the system some headroom, those usage amounts vary depending on what you are doing etc.

Another observation is drop frames /stuttering or lag seems to occur when either GPU or CPU reaches 100%. GPU is obvious but CPU it should be the same with AMD CPU you'll find one core pretty much doing all the work so CPU usage might say 40% for a 4 core however you need to look at the core which is most active and keep that one from hitting full utilisation as well.

If you haven't already discovered all the maps preform slightly different keeping settings constant and observing CPU and GPU usage with the different maps for me NTTR is the least GPU/CPU intensive map.

NTTR pretty much always 45FPS

Caucasus slightly more demanding slightly more frequent FPS drops and occasional lag

PG more regular drops below 45FPS and sometimes lag

Normandy somewhat more regular drops in frame rate and lag

HTH

The shader mod gives a bit of a boost to the GPU for increased PD. https://forums.eagle.ru/showthread.php?t=215373

 

I run 1.4 PD on high settings (bar shadows-flat) with 4x MSAA on a OCèd RTX 2080 & i7 4790k 4.6ghz, 24 gigs ram. The main issue is cpu optimization is poor and chokes the GPU when the map is busy. Again anything from 45-90fps. Usually 90fps when at altitude 1000ft+. Treetops 45-50fps. Nevada map can hover around the 90fps close to the ground.

Brilliant! I don't know squat about computers, know a bit about the brain though. I think we want edge detection in the center of vision, that's what central vision does. Applying MSAA to central vision is the shiznit. Don't need or want it everywhere else. Guy's a genius.

The brain models objects and environments. Up to this point computers did a great job of helping us model objects, which is hard. VR doesn't model objects very well, but it models environments very well. Which isn't hard at all, it's the most natural thing in the world. The most direct connection between the 3D model in the machine, and the 3D model we are all trying to build in the mind.

Since you mention it, the Nevada map has real potential IMO. I wrote a dystopian/WWIII novel a couple years ago, the Nevada map would be the ultimate sequel. Believe it or not it's the most logical setting given how I left the thing. Like, I think I'm gonna write it chapter-by-chapter in DCS multiplayer missions, Clancy-style. Ain't about rendering grass, ya know? Setting up the server as we speak. Next the website. Next up are the Twitter accounts for CINC WTO, and Radio Free Mexico, and the Chinese Humanitarian Mission (gotta hand it to

'em, they bring assets). The photo contest, because I'm getting reports there are twin-tailed aircraft flying training missions south of Laughlin, and I need to know if those are MIG-29's or SU-33's, and it matters, because that day is coming, and we need to be ready. Follow me?

Point being we aren't rendering an image, we are rendering a world model. We need to be looking in the right places to make that happen. It's only sort of about how well you can read the altimeter, you know? We need to knock out the technical stuff and get on with it IMO. We have bigger fish to fry. My theory is, if you set up a server for virtual reality, you'll get it.

Anyway yes, thank you. There's a lot of content in your reply and I'm listening.

Brilliant! I don't know squat about computers, know a bit about the brain though. I think we want edge detection in the center of vision, that's what central vision does. Applying MSAA to central vision is the shiznit. Don't need or want it everywhere else. Guy's a genius.

 

The brain models objects and environments. Up to this point computers did a great job of helping us model objects, which is hard. VR doesn't model objects very well, but it models environments very well. Which isn't hard at all, it's the most natural thing in the world. The most direct connection between the 3D model in the machine, and the 3D model we are all trying to build in the mind.

 

Since you mention it, the Nevada map has real potential IMO. I wrote a dystopian/WWIII novel a couple years ago, the Nevada map would be the ultimate sequel. Believe it or not it's the most logical setting given how I left the thing. Like, I think I'm gonna write it chapter-by-chapter in DCS multiplayer missions, Clancy-style. Ain't about rendering grass, ya know? Setting up the server as we speak. Next the website. Next up are the Twitter accounts for CINC WTO, and Radio Free Mexico, and the Chinese Humanitarian Mission (gotta hand it to

'em, they bring assets). The photo contest, because I'm getting reports there are twin-tailed aircraft flying training missions south of Laughlin, and I need to know if those are MIG-29's or SU-33's, and it matters, because that day is coming, and we need to be ready. Follow me?

 

Point being we aren't rendering an image, we are rendering a world model. We need to be looking in the right places to make that happen. It's only sort of about how well you can read the altimeter, you know? We need to knock out the technical stuff and get on with it IMO. We have bigger fish to fry. My theory is, if you set up a server for virtual reality, you'll get it.

 

Anyway yes, thank you. There's a lot of content in your reply and I'm listening.

What kind of missions do you guys play with VR? Multiplayer, design your own SP? Do you not find the campaign DCS SP content to be too taxing generally?

I want to play the epsom and WW2 campaigns but when the flak opens up it gets way too stuttery, and basically theres just too many units! Looks like an amazing campaign so it really pains me!

Damned shame the people who design these missions don't take VR users into consideration but I get we are a very small propotion of the sim's base.

Anyone know of, or have any WW2 missions to share, that work well in VR?

I can make my own but its too predictable to play ones own missions all the time, and I don't have much time to make them anyway.

I do not need much, I'm happy basically with fly up, out and engage, and then fly back, with maybe a bit of ground attack potential.

One thing I have noticed in VR as well that missions must have all assets loaded into them at mission start, otherwise there will be nasty stutters and FPS drops when these assets load in mid mission. With DCS thats probably good advice for all user with the MB anyway I suppose...

What kind of missions do you guys play with VR? Multiplayer, design your own SP? Do you not find the campaign DCS SP content to be too taxing generally?

 

Damned shame the people who design these missions don't take VR users into consideration but I get we are a very small propotion of the sim's base.

Thats why im starting up a server, to cater to vr enthusiasts. I have this grand vision of a story driven campaign but the main thing I think will be collaboration on mission design, what kind of missions are the most fun in VR, that leverage The tech the best kinda thing.

Unfortunately with VR we are all mid-level GPU's!

What does that mean?

What does that mean?

It means that whether on a 1070 or 2080 ti you'll still be ball's deep into reprojection.

The lowest level of shadows, think it's called flat, don't really have a hit, but very useful for spotting fast movers low. I only see them under active objects, like flying craft, statics, ships, and vehicles. The simple terrain shadows seems to do the same for buildings in cities, anything else on mine for shadows I'll know it.

Can't prove it, but suspect setting SS outside the game in steamVR also applies it to the desktop image, because it sure seems a greater strain for the same amount.

Is it possible for MSAA settings to be fractional? The 2 setting seems a bit too much, if it were possible to have half or a third I'd be happy. Only need it to fix the thin straight objects like high line poles, or airfield towers, the mast assembly on the carrier, they all look horrid with bright pixels flashing. Otherwise it's too big of a hit even at 2. Once I get off the ground it's all OK.

It means that whether on a 1070 or 2080 ti you'll still be ball's deep into reprojection.

Roger that, we are seeing performance tank with current technology offering only marginal improvement at this time. :thumbup:

Roger that, we are seeing performance tank with current technology offering only marginal improvement at this time. :thumbup:

Well, I don't know if I agree. I saw a big improvement on my Vive Pro when I upgraded my 1080 Ti to a 2080 Ti last week.

Well, I don't know if I agree. I saw a big improvement on my Vive Pro when I upgraded my 1080 Ti to a 2080 Ti last week.

Wonder the extent to which us mid-level types might be better off upgrading CPU. I figure there's only so much Oculus can do, kinda saving my ammo to see what (if anything) will dive 4k VR headsets.

My theory is, there's a point of diminishing returns. You got your AA settings, which I figure is how hard you're flogging the GPU. That leads I think to reasonable compromises in game settings, things like visible radius, that may not make much difference given the limitations of VR, so we can ease up on CPU load. And then finally, tweaks to mission design. All predicated on pixels per degree basically

Well, I don't know if I agree. I saw a big improvement on my Vive Pro when I upgraded my 1080 Ti to a 2080 Ti last week.

But are you still in reprojection most of the time? I used to OC my CPU to 5ghz cause I thought it was giving me big gains, but one day found I had disabled it and hadn't even noticed!

What are the numbers FPS wise with the 2080TI compared to the 1080Ti in DCS. Say in normandy down low or something?

But are you still in reprojection most of the time? I used to OC my CPU to 5ghz cause I thought it was giving me big gains, but one day found I had disabled it and hadn't even noticed!

 

What are the numbers FPS wise with the 2080TI compared to the 1080Ti in DCS. Say in normandy down low or something?

I don't think CPU is an issue in DCS. In fact I'm not sure what the issue is. When running DCS in VR, sometimes fps struggles at higher settings, but the GPU is never over 70% usage, which suggests to me, at least, that something isn't right.

When I run in 4K on my monitor, I limit fps to 58 to get Gsync all the time, and I run at max settings. The fps never drops below 58. In addition, the GPU is never really stressed. If I take the frame limiter off, then the fps is usually over 100, but with all the issues that has on a 60Hz monitor.

So, why is VR so (relatively) bad?

In DCS VR, I get about 50% reprojection rate, which is not great, but again, this seems to be a DCS issue as running other games I don't have that. Skyrim VR, for example, tends to run at or near 90 fps, and the reprojection rate is negligible, around 0.4%.

Regarding your question about fps, it's difficult to say, as settings are critical, etc., and I can't remember what settings I used to use with my 1080 Ti. Sorry!

I don't think CPU is an issue in DCS. In fact I'm not sure what the issue is. When running DCS in VR, sometimes fps struggles at higher settings, but the GPU is never over 70% usage, which suggests to me, at least, that something isn't right.

When I run in 4K on my monitor, I limit fps to 58 to get Gsync all the time, and I run at max settings. The fps never drops below 58. In addition, the GPU is never really stressed. If I take the frame limiter off, then the fps is usually over 100, but with all the issues that has on a 60Hz monitor.

So, why is VR so (relatively) bad?

In DCS VR, I get about 50% reprojection rate, which is not great, but again, this seems to be a DCS issue as running other games I don't have that. Skyrim VR, for example, tends to run at or near 90 fps, and the reprojection rate is negligible, around 0.4%.

Regarding your question about fps, it's difficult to say, as settings are critical, etc., and I can't remember what settings I used to use with my 1080 Ti. Sorry!

Not sure what benefit your 2080Ti has over 1080Ti and I would be interested to know, however as for DCS if you look at CPU performance you should see one core pretty much hammering along at near 100% depending on settings etc.

As for why DCS has difficulties with VR I suspect it's to do with code efficiency and shaders, given that in DCS ver 2.0..2.3 I could get constant 90FPS with a 980Ti under VR now I struggle to maintain 45FPS with a 1080Ti on PG and Normandy maps. That's progress for ya. :cry:

Not sure what benefit your 2080Ti has over 1080Ti and I would be interested to know.

Really don't know what you mean. The 2080 Ti has produced a big improvement over the 1080 Ti in DCS. Increases in fps and the ability to ramp up the settings. What more do you need to know?