What are the best settings for maximum performance?

[rtschramm]

I have an I9-7900X running at a base speed of 3.31 GHz, with 64 GB memory running at 2934 MHz , a 2TB M2 SSD, and an overclocked 2080 it with 11 GB of memory. I am running the game through Steam VR with a valve Index.

I set the mirror screen to 640 * 480 windowed mode, and then I minimize it to not use waste any CPU time rendering to the computer monitor. I have the game's executable in the turbo-boost list. The CPU does turbo-boost at 4.3 while playing the game.

I cannot get this game to running acceptably unless I run the game at 80Hz and turn the screen resolution down to 100% for both the application and steam VR. I have motion smoothing and super sampling turned off. The problem with this configuration is that the screens is not as sharp as when I play the game at 90 Hz and bump up the resolution to 120% on the steam resolution property.

I have not touched the NVidia control panel yet.

According to the Steam VR performance test, my computer system is well above the specifications to play VR games.

Note that I can play Half-Life: Alyx with every setting turned up to high and I experience no stutter or jerkiness when moving.

So, am I am missing something? I am using the latest beta version because I experienced the same problems with the stable version.

What Nvidia settings are you using to get this game to run smoothly. Note that things get bad when I play a mission with many AI players.

I really like this game, but with the performance issues, it is getting almost unbearable to play this game.

[Slice]

I have fixed the performance during heavy missions on my Ryzen 3900X (this would work with Intel too). I had ran numerous test to analyse the performance data, comparing both cases, before and after, and the results are positive, this method works perfectly.


The problem is that SteamVR has many components that work in the background to make the VR magic happen. Those SteamVR threads sort of "fight" with Vtol VR to monopolise the CPU cores time and causes a mess that spikes the latency resulting in bad VR performance (retroprojection, smoothing etc) . This issue is caused "mainly" by the Unity Engine multithreading ♥♥♥♥ show, and "lack" of Vtol VR final optimization (Comprehensible, still in development). *** I believe physics are still not threaded***

The main objective here is to separate the SteamVR processes (mainly "vrmonitor" and "vrcompositor") from the game threads. Furthermore, AMD unique architecture is also to be considered for Ryzen users.

*Ryzen users:

Allow me to explain something. Some Ryzen CPUs have more than one CCD (chiplet design). In the case of the 3900X, there are two CCDs, each one contains 6 cores with 2 logical threads each, giving a total of 24 logical cores.

CCDs are separated from each other and connected by the infinity fabric (intel calls this technology "glue")

If one process would use 2 logical cores from CCD 1 and 2 logical cores from CCD 2, there would be time spent for the information to travel from CCD 1 to CCD 2 and so on. Most of the time, it would be much more logical and efficient to keep that process within one CCD if possible, specially when latency is crucial! But the windows scheduler dont know any better so we are here to help!

Vtol VR and Ryzen 3900X:

In this case I would choose SteamVR and other steam related processes to work in CCD2 and Vtol VR in CCD1, making this a strict rule and forcing it (No windows, today you do exactly what I tell you)

This would look like this:

CCD 1
________________________________________________________

C0 - Vtol VR
C1 - Vtol VR
C2 - Vtol VR
C3 - Vtol VR
C4 - Vtol VR
C5 - Vtol VR
C6 - Vtol VR
C7 - Vtol VR
C8 - Vtol VR
C9 - Vtol VR
C10 - Vtol VR
C11 - Vtol VR
________________________________________________________



CCD 2
________________________________________________________

C12 - vrmonitor, vrcompositor, vrdashboard, vrwebhelper
C13 - vrmonitor, vrcompositor, vrdashboard, vrwebhelper
C14 - vrmonitor, vrcompositor, vrdashboard, vrwebhelper
C15 - vrmonitor, vrcompositor, vrdashboard, vrwebhelper
C16 - vrmonitor, vrcompositor, vrdashboard, vrwebhelper
C17 - vrmonitor, vrcompositor, vrdashboard, vrwebhelper
C18 - vrmonitor, vrcompositor, vrdashboard, vrwebhelper
C19 - vrmonitor, vrcompositor, vrdashboard, vrwebhelper
C20 - vrmonitor, vrcompositor, vrdashboard, vrwebhelper
C21 - vrmonitor, vrcompositor, vrdashboard, vrwebhelper
C22 - vrmonitor, vrcompositor, vrdashboard, vrwebhelper
C23 - vrmonitor, vrcompositor, vrdashboard, vrwebhelper
***(I put all steam processes but it is not necessary)***
________________________________________________________


Vtol VR would be contained within CCD 1 and all Steam VR processes would be contained within CCD 2. This would one; improved the latency for Vtol VR and the Steam VR compositor by stopping the threads from constantly jump from one CCD to another and second; stop the threads from fighting to monopolise cores (increasing latency again)

***I also set Vtol VR, vrmonitor and vrcompositor to high priority processes, from normal***

***Remember to strictly enforce the affinities***

NEW *** I am also getting great results disabling SMT from the Vtol VR process (you manually force this by omitting the odd cores (disable C1, C3, C5, C7, C9, C11 from the Vtol VR affinity). I could shave off between 1 and 2 median ms from the CPU latency during high load missions*** NEW

***AMD Threadripper users would have to take into account the NUMA nodes***


With an intel, lets say, an hexacore for example, I would give to Vtol VR 4 cores (8 logical cores) and 2 cores (4 logical cores) for steam processes.

The results are "shocking", see for yourself. Spikes are GONE. This method would work for many other games or case scenarios, not only VtolVR.

The tool that I use for setting up all of this is "Process Lasso Pro", the tool I use for checking my CCD configuration is "Ryzen Master".

Process Lasso has a free version for trying it out.

If anyone needs help trying this out, let me know.

[Stalker]

I don't have a clue what you just said. But if you've found something to optimise the game for the rest of us how about you tell the Dev ?

[Slice]

引用自 Stalker：

I don't have a clue what you just said. But if you've found something to optimise the game for the rest of us how about you tell the Dev ?

The dev isn't responsible for the windows CPU scheduler or the poor unity multithreading optimization.

[rtschramm]

I also tried Lasso Pro, and you are correct if you separate the steam VR processes from the other VR programs, but is too complicated for the average use to do, plus this requires using a serious program that can screw up your system system performance.

On my system, Lasso Pro didn't give me the same performance as just disabling hyper-threading in the bios and also disabling threaded optimization in the Nvidia control panel.

I had another issue with Lasso Pro where editing in Unreal Game Editor was impossible, so I had to uninstall it.

[rtschramm]

My fix cannot be done by the developer, it must be done on an Intel processor that has hyper-threading enabled. You have to reboot into the BIOS and change a setting. NOTE that this is not that difficult, but can also screw you computer up, if you disable the wrong setting.

[_provector]

Hey dude, I feel ya. I've been with the game since the beginning and depending on the current update the performance is ups and downs. The biggest bottleneck seems to be the cpu at the moment. I can actually crank up supersampling and other graphical tweaks as gpu is barerly 60% busy. The 1.0 seems to be hapenning this year so let's hope for a wave of optimizations.

[_provector]

hmm, i just played the game again and have smooth 90 fps, the problem occurs on a bigger missions with huge maps. Do you have multi monitor setup? If so disable any extra monitors and try running the game. I always do that in vr as on certain systems (mine) multi monitors cause huge frame drops in vr.

[Burnit2EarnIt]

I am running i7 4790k 4.0 NO OC
gtx 980
16 go ram
Turned my frame rate to 70
But the biggest thing I saw improvement was changing codec to h264.
play multiplayer with father and brother all in a game the biggest thing is when you get close to another vehicle online playing with you it shutters a little bit but not unbearable.
Not sure why guy with a super computer is having issues lol.
Lower frame rate with it not being that detailed anyway in vtol compared to like fsx with add one being super realistic. Should have no issues.

My specs way below most

[_provector]

引用自 liveitupcali：

I am running i7 4790k 4.0 NO OC
gtx 980
16 go ram
Turned my frame rate to 70
But the biggest thing I saw improvement was changing codec to h264.
play multiplayer with father and brother all in a game the biggest thing is when you get close to another vehicle online playing with you it shutters a little bit but not unbearable.
Not sure why guy with a super computer is having issues lol.
Lower frame rate with it not being that detailed anyway in vtol compared to like fsx with add one being super realistic. Should have no issues.

My specs way below most

Most likely your CPU is the bottleneck here. The game is very CPU intensive, less heavy on the GPU side. 4 cores are in most cases not enough. I went from my i7 6700k to ryzen 5900X with same graphics card and it was a huge difference.

[Cereal Killer]

引用自 rtschramm：

Once I learn how to play DCS World in desktop mode, I plan to play it in VR mode to see how it compares to VTOL VR as far as framerate goes. DCS World uses its own proprietary game engine. DCS World is way more complex that VTOL VR.

DCS performance depends a lot on the module (aircraft) you are using. For example the JF-17 module works beautifully for me but the AH-64D helicopter stutters like crazy and was virtually unplayable until I switched over to the new multi-threaded beta version.

I have a pretty robust system. 13600k, 32Gb RAM, and a 3080. Even with all of that, I still have to make a lot of tweaks and compromises with DCS. A lot of folks have been CPU-bound for years and years even with high-end CPUs because the game has been single-threaded up until about a month ago when they finally released the multi-threaded beta. I personally have experienced a good 20fps increase in a lot of situations which has made it much more enjoyable but there is still a lot of progress to be made. Not everyone has experienced big gains like that (for a variety of reasons) but the multi-thread version is still in its infancy and will almost surely improve as it ages and gets worked on. The devs are also supposed to be implementing Vulcan pretty soon which a lot of people believe will further streamline the rendering pipeline and increase frames even more.

[_provector]

Smaller the map you will be playing on, and less units in the mission means more performance. Try some really small maps, I did that on my old pc. Game is really CPU bound, not that heavy on gpu side.