Upcoming updates, MT, DLSS, all the goodies.

[SPAS79]

Hello, 

BLUF: discussion on the elements in the topic is usually full of good information that applies to DCS and PC gaming in general, and I'd like to stimulate it in a dedicated space (like: this post) to keep learning how that will apply to DCS and what that is in general. 

I'll start with what I think, but a disclaimer first: I fly only in VR so my observations and considerations are from that angle.

So here goes: 

MT will help somewhat, especially in CPU heavy scenarios but it will probably need more RAM to work properly and will benefit from fast GPUs that can keep up with more efficient CPU work. 

If and when we get DLSS, that might help too but we will need to strike a balance between definition and fps. I am using FSR now with OpenXR toolkit and while it does help to a degree, you have to carefully tune it to keep instruments and MFDs readable.

Vulkan has the potential to bring the biggest improvement as I understand the API allows for a much lower level access to the rendering hardware, reduces overhead, and can enable more efficient rendering in the case of VR (If I am not mistaken, for VR Vulkan can render "differential" sections that then get added to the common frame to form the two projected images for VR headset, so it renders the common portion of the frame and then the parts that are different for left and right, cutting down time as it doesn't need to render the full scene twice). Looking at how it was implemented in X-Plane 11 tho takes some optimism away as it kind of sort of made everything a little smoother over OpenGL, but just a little. But I am sure ED is better than that. 

This thread exists because I have seen some threads being locked because the discussion invariably ended on MT, DLSS and the supposed performance gains that have either been hinted at on various discord channels by developers collaborating with ED, or inferred from the information teased in some official DCS video illustrating various upcoming aspects (all not related to the performance of the core sim engine but about various systems implemented in the modules - i.e. the new ARC-210 radio for the A10C II). 

I have been enjoying the discussions quite a lot, not because I like to read speculative hypotheses on how this or that will be better or worse, but because within the discussions there usually are bits of information that I didn't know about how the rendering pipeline works, what MT actually does, how DLSS functions, how to gauge CPU/GPU usage in DCS and so on. 

So, hopefully this thread will host all of this discussion. 

Thanks for reading my blog. 

EDIT: this should really be in DCS Core Wish List - ED Forums, but I didn't know it existed until this edit. 
EDIT #2: the thread has been moved to Chit-Chat? Ok, I guess? 

Edited March 10, 2023 by SPAS79

[Zebra1-1]

Good thread/topic. 

Your post seems consistent with other articles/posts I've read on this. I don't think MT will bring the performance that some people are hoping for, but it seems like it could improve over time. I'm honestly not knowledgeable on this subject though. As I understand (and could be wrong), MT will be enabled via Vulkan in DCS. So I assume that whichever patch brings MT, will have Vulkan in some form or another. If I were to guess though, DLSS will likely come with the full Vulkan implementation (or the graphics side of it). If that even makes sense.

[Dragon1-1]

No, MT will come first, and if you hop over to the other thread, you'll see people (3rd party devs and closed beta testers) getting absolutely massive gains from it. The plan is quite elaborate, and DLSS is pretty far down the line, with Vulkan even further down.

[Mr_sukebe]

1 hour ago, Dragon1-1 said:

No, MT will come first, and if you hop over to the other thread, you'll see people (3rd party devs and closed beta testers) getting absolutely massive gains from it. The plan is quite elaborate, and DLSS is pretty far down the line, with Vulkan even further down.

Were those massive gains in VR or on a monitor?

DCS in VR “appears” to be more GPU bottlenecked.  If that is the case, the benefits from multi threading may well be substantially lower

Edited March 10, 2023 by Mr_sukebe

[Dragon1-1]

Both. 

VR is GPU-bound because DX11 renders the whole scene twice, and as such, we're trying essentially to push something like 2x4K pixels, which is a lot. This appears to boost overall rendering efficiency, with proportional gains to be expected on both monitors and in VR. 

[SkateZilla]

1 hour ago, Mr_sukebe said:

Were those massive gains in VR or on a monitor?

DCS in VR “appears” to be more GPU bottlenecked.  If that is the case, the benefits from multi threading may well be substantially lower

 

VR is very much just as CPU Bottlenecked as 2D.

Considering the CPU Bottleneck is caused by Draw Calls and Object counts, and when in VR everything is called/Rendered 2x...

[SPAS79]

So this is what I get with a quick and dirty test at 1080P (2D pancake), eyeballing the avg fps in the very first 30 seconds of flight (with no input from the user).
PC is 13600K5.5GHZ all P Cores, 4.3GHz E-Cores, 64GB DDR4 @ 3600, RTX 3090 with modded bios (power limit to 470W)

I'd say it's a net improvement, I will start using it in VR and see if there's any benefit there too.

F18 - Syria Free Flight Instant Action

 FPS avg  |           Version        | Shadows | Terrain Obj Shadows
   160    |  DCS 2.8.2.35759 (no MT) |  High   |        Default
   210    |  DCS 2.8.2.35759 (no MT) |  Off    |          Off
   230    |  DCS 2.8.3.37556 (MT)    |  High   |        Default
   275    |  DCS 2.8.3.37556 (MT)    |  Off    |          Off
   265    |  DCS 2.8.3.37556 (MT)    |  High   |          Flat