VR Performance in DCS World

[DerekSpeare]

I went into my NVIDIA control panel and vsync was enabled for global profile (not adaptive, I don't see the option). I turned it off and it doesn't appear to have any effect.

 

As for GPU clocks, using MSI Afterburner, my Core clock goes to 975 when Oculus is launched, then 1400 when DCS is launched. So I don't think that is the problem.

 

Any other suggestions?

 

What does your clock speed do when you're actually flying?

[FalconFSX]

What does your clock speed do when you're actually flying?

 

Mostly stays right around 1400 with minor bumps no lower then 1300 for a short period. The GPU usage % is moving a fair amount, 40-65%.

[wasyl00]

Just to confirm. I'm having low frame rates too. Similar to you guys I have fairly strong rig at the moment 6700K @4.7, 980Ti @1500 and experiencing performance issues in 1.5 and also NTTR as well. I set everything to low and still frames drop quite frequently to 45fps which is the result of ATW kicking in. I really was hoping to get at least stable 90 with this setup but nope. The biggest pain IMHO is to observe how excellent and fluid everything is at 90 and how slightly blurry it goes when the performance drops. Don't get me wrong, I love how ATW still keeps the head movement judder free but I wish that the framerates were a bit better. Hopefully ED is still tweaking the engine so we can get some performance boost.

[j0nx]

If Gpu usage is not pegged at 99% and cpu at 40% then something is wrong. Either ED has some serious issues to work out with VR and their game engine or maybe you have vsync turned on.

[DerekSpeare]

Frame rates are locked to the VR headset's refresh rate, effectively creating a vsync condition; it's 75 for the DK2 or 90 for the Vive or CV1. This is a consequence of VR. There is no way to change it, and one would not want to do that anyhow.

[vicx]

If Gpu usage is not pegged at 99% and cpu at 40% then something is wrong. Either ED has some serious issues to work out with VR and their game engine or maybe you have vsync turned on.

 

This might be a partially valid claim but I would love some evidence. Do you have graphs and video of VR perf in other applications and games that show this to be the case? Do you have evidence in games where ATW is being activated to show that this is the case. Sure in an ideal world GPU and CPU would be fully utilised but with present GPU and CPU architectures I think this VERY HARD to do with VR.

 

When ATW is activated, frames are discarded which reduces GPU load and in engines where framerate and simulation rate are linked (MOST GAMES and APPS) then CPU load should also drop. Blame current hardware/software architectures if you do not like this.

 

 

---

 

Other point that that ED has some work to do is an overstatement and understatement. We can't say that ED is doing anything wrong with VR DCS until we can point to a competitive simulation software that is doing VR better. Currently there is no other software in it's class doing better. Saying ED has work to do it is like saying "we need technical progress". Of course this statement is ALWAYS true and one wonders how often it needs to be repeated.

 

We can also say that Nvidia and AMD have a lot of work to do because we need technical progress for GPUs too. 980TI was just a stop gap measure when it comes to VR compatible GPUs using current GPU architectures. The hardware architecture of a 980TI is not ideal and future architectures will show this. Further to this, currently deployed and mature APIs (DX11) are not ideal for VR. Brand new APIs (DX12) are cutting edge and supporting them is very expensive and will take a long time for ED.

 

---

 

So for those who are not good at waiting, what can be done while waiting for technical progress to arrive? Maybe satisfy your curiosity and perhaps identify small gains through trial and error. I did a lot of testing last year to try and improve perf in NTTR/Vegas by culling objects out of the map and modifying (like a caveman with a laptop) some shader code. I wasn't worried about how ugly or silly things looked - I was only looking for more perf. Anyone curious can do the same.

 

How to record and share testing is difficult. Last year screenshots and video taken in VR in DK2 did not show debug info. I have noticed that SteamVR has better tools for this kind of thing. When I get my Vive (MAY :noexpression:) I will likely do some testing with it.

Edited April 23, 2016 by vicx
Lets do testing and show screenshots and evidence for our claims

[BigDuke6ixx]

^ well said.

[DerekSpeare]

I turned off my Oculus and cranked up all the detail in DCS to max...forgot the AA and it was 2x. My GPU utilization never went above 66%, and that was only when zoomed in at distances on the horizon - no other objects other than the F15 were present (building, of course). My GPU % stays around 36% with the Oculus on.

[SkateZilla]

THis would Pretty much sum it up:

 

[BigDuke6ixx]

I turned off my Oculus and cranked up all the detail in DCS to max...forgot the AA and it was 2x. My GPU utilization never went above 66%, and that was only when zoomed in at distances on the horizon - no other objects other than the F15 were present (building, of course). My GPU % stays around 36% with the Oculus on.

 

Yeah, something not right there. In Elite: Dangerous the GPU hovers around 95%

[SkateZilla]

Yeah, something not right there. In Elite: Dangerous the GPU hovers around 95%

 

Flying in Space w/ next to no objects vs flying over a terrain w/ 100,000+ Objects.:music_whistling:

[BigDuke6ixx]

Flying in Space w/ next to no objects vs flying over a terrain w/ 100,000+ Objects.:music_whistling:

 

Never had a problem in space even with my old computer. However, with my new systems, I saw a massive increase in performance on planets (lots of terrain, other wing members and huge bases) with OH 1.3 and ATW.

[SkateZilla]

Never had a problem in space even with my old computer. However, with my new systems, I saw a massive increase in performance on planets (lots of terrain, other wing members and huge bases) with OH 1.3 and ATW.

 

I believe Elite: Dangerous uses a different mesh rendering for the terrain, and they still dont have no where near the number of objects on the planet surface that ED Does on the Terrain.

[RUS38]

Flying in Space w/ next to no objects vs flying over a terrain w/ 100,000+ Objects.:music_whistling:

 

Ok, so what you are trying to say is that GPU usage will drop as you add more objects? Still doesn't sound right.

[SkateZilla]

Ok, so what you are trying to say is that GPU usage will drop as you add more objects? Still doesn't sound right.

 

Ive explained the DirectX API CPU Overhead 100x over already, Im not gonna do it again.

 

Basically, More Objects = More DirectX API Draw Calls that have to be processed by the CPU, the more draw calls that flood the CPU the more the CPU gets backed up, the more the CPU gets backed up the les intructions the GPU receives, which causes the CPU to sit idle. Hence low GPU Usage.

[vicx]

Ok, so what you are trying to say is that GPU usage will drop as you add more objects? Still doesn't sound right.

 

Only someone with an intimate knowledge of the DCS engine can say for sure. But with only a limited knowledge of DCS rendering I can see how this "could" be true in some circumstances.

 

DCS uses instancing (lots of objects) and other techniques in their DX11 engine to maintain acceptable perf on a fairly wide range of hardware. It could be (I do not claim to know for sure) that their techniques were chosen and tuned with a target of 30fps-60fps in mind. This makes sense because before VR arrived the expectation was that a new DCS engine should drive a triplescreen or a 4K screen with as many objects on screen as possible, with as much detail and as much visual fidelity as possible; with 60fps desired and 30fps as acceptable. What people wanted most was high resolution with a big draw distance (you get a LOT of objects with the draw distance).

 

The main point is that ED wasn't optimising for a constant 90fps. To be fair AMD and Nvidia hardware and drivers weren't optimised for 90fps either. The goal in hardware and software for years and years has been maximum visual quality at 60fps using every trick in the book.

 

Now with VR we want a constant 90fps. Even if we want to trade away long draw distances and have fewer objects to get 90fps it isn't as easy as you think. The optimisation at every level of software and hardware is to deliver 60fps ... for all hardware and software it is going to take a while to unravel existing optimisations and create parallel optimisations for VR rendering purposes.

 

With this in mind we can do our testing ... we may see counter-intuitive results. I won't be surprised if we do.

[Milopapa]

Vive settings on 2.0.2?

 

I'm struggling to get above 30fps in 2.0.2 with minimum settings on a 980Ti and i5 4.2GHz with 16GB RAM.

 

Can anyone tell me how to achieve a usable framerate on a Vive?

 

In 1.5, I can get 45fps (drops over populated areas though and rarely reaches 90fps).

It doesn't even get close to maxing out my CPU (40-50% max CPU), and my GPU is sitting at 30% load. Something's not right.

 

I get that nobody optimized for VR before and it's computationally intensive etc etc but I just want to fly this thing on my $2000 rig.

[j0nx]

I noticed yesterday that my GPU on DCS with surround was only at about 60% utilized on mostly high settings. I'm positive in regular surround that it was pegged at 99% before these last updates. I'll have to try 3D surround which I know should be pegged at 99%. My fps has dropped a lot with these latest updates too but there are a lot of new features now that didn't exist then so... My old 2600k has been offline for a few months and hasn't gotten any of these updates. I am going to fire it up this week, copy over the old 1.5 and 2.0 directories to my new box and check gpu utilization. I'm starting to think that maybe these latest patches have decreased cpu and gpu utilization causing performance issues for people with higher end setups. Anyone else seeing this?

[SkateZilla]

I'm struggling to get above 30fps in 2.0.2 with minimum settings on a 980Ti and i5 4.2GHz with 16GB RAM.

 

Can anyone tell me how to achieve a usable framerate on a Vive?

 

In 1.5, I can get 45fps (drops over populated areas though and rarely reaches 90fps).

It doesn't even get close to maxing out my CPU (40-50% max CPU), and my GPU is sitting at 30% load. Something's not right.

 

I get that nobody optimized for VR before and it's computationally intensive etc etc but I just want to fly this thing on my $2000 rig.

 

You should be getting more than that, alot more than that actually,

 

I get 30 -45 Fps on a 7870 w/ 2GB in rift w/ everything but textures on Low.

 

Post a screen of your graphics menu.

 

And your DCS.exe or nVidia ControlPanel/Inspector Settings.

[Milopapa]

I did some testing/troubleshooting last night.

With the latest alpha, I was struggling to get decent fps even in 2D:

<screenshots would go here, but Imgur refuses to show me my images>

All low settings, flat shadows etc - it was giving me ~70-80fps sitting on the runway in a Huey in 2D. Same settings 22-30fps in VR.

 

I used to run the Alpha at stable 60fps with my previous graphics card (R9 280X) with mostly low-medium settings so this raised a flag.

I ran 3DMark, everything's fine.

I decided I'd try reverting to older version of the Openalpha and it finally did the trick!

I downgraded to 2.0.0.49135, and it was giving me 110-120fps with the same settings (vs 70-80 in 2.0.2) and this translated to a stable 45fps in VR.

I'm now able to fly over Las Vegas and maintain the 45 fps in the Vive.

I even upped some settings from Low, like 4xMSAA and I'm still good.

GPU is still sitting around 35%, CPU is ~50%, so there is still ways to go optimizing this, but at least it's not a stutterfest.

 

I really wish ED would do something about their thread management, it's crazy that it can't utilize high-end hardware properly. Reminds me of Arma.

[wasyl00]

GPU is still sitting around 35%, CPU is ~50%

 

 

Check your CPU performance by each thread. I'm pretty much sure one thread will be pegged at 95-98%. My overall CPU usage was around 25% until I checked this. DCS is really hard on CPU so my 6700K is really struggling @4.8 (one thread at 98%) atm while 980TI sits at around 60%.

I'm able to get pretty stable 90fps with everything on low but only in FC3 planes sadly.

 

Still low pass over Vegas in F15 knocks my socks off even at low settings.

[Milopapa]

Check your CPU performance by each thread.

Last time I checked, the load was spread across all 4 cores, but I'll double check when I'm home.

 

What's bothering me is, if both CPU and GPU are underutilized, what's bottlenecking the game? If it's really the DirectX API calls, shouldn't ED consider moving down either the DX12 or Vulkan path? I know it's not that easy, but the benefits could be huge.

[SkateZilla]

1. They arent even finished w/ DX11 lol

2. DX12 SDK is not in Open Distribution

[dot]

1. They arent even finished w/ DX11 lol

2. DX12 SDK is not in Open Distribution

 

https://blogs.msdn.microsoft.com/chuckw/2015/08/05/where-is-the-directx-sdk-2015-edition/

 

"There is something called the DirectX 12 SDK which was used as a beta vehicle for the development of DirectX 12 through the Early Access Program. Now that the Windows 10 SDK is final, you don’t need access to the DirectX 12 SDK at all."

[SkateZilla]

Well that explains lack of updates on my SDK, lol. But I havent been pre-occupied w/ DX12 in a while.

 

I Actually have the Windows 10 SDK, just never used it, as the market share is still not anywhere near the % I'd consider alienating Windows 7 and 8 Users.

Edited April 28, 2016 by SkateZilla