[REPORTED]2.5.6 Performance

[squirrelmeister]

So I bit the bullet on TGW's server and rolled us all back to stable and we all instantly noticed almost a 20 - 40% increase in performance on our machines measurable in both Monitor and VR, but not only that there is a significant increase in the server side reported FPS if your using something like Tacview to monitor said performance.

 

Here's just some things we noticed performance wise over the last 4 weeks or so... Prior to the SC dropping we saw server frames of about 70-90fps base line dropping over time down to about 60fps with a lower limit of 40 on our 'heaviest' mission at the time. (Red Iberia). When what is now 'stable' dropped that jumped with 119/120 being the 'baseline' fps and the lowest being 70fps. At the same time on my own machine I was seeing 40-100fps prior to stables release but often with extremely large frame time 'spikes' on monitor and well VR wasn't useable half the time as bowwave would crash out. With SC release frame rates went back to 60 - 120fps monitor and 30 - 60 fps on VR. There is still almost a 'stutter' in frame times every 12-20 seconds but it's not a 'major' stutter..

 

The releases in open beta after SC however saw all that drop rapidly again, I don't know if it's the radar, if something is messy with AI or what but something is defiantly causing a perf hit that impacts the GPU cycles on machines running GPU but also the reported 'frame rate' on servers.. as at times our server frame rate was falling to as low as 24fps and we were actually forced to stop running 1 mission all together due to the performance hit. Worse the frame times are not anywhere near smooth which makes drops in frames even more noticable.

 

Wednesday we rolled back to Stable .. and with out any changes to the missions we are running instantly we are back at the 100+ fps rates on the server and the smoother frame rates in clients.

 

If I was a gambling man I'd almost place bets on it being

A. The Radar code changes.

B. The lighting code changes.

C. fixes to AI Routing.

 

I'd lean more towards A and C given that C has had this type of impact before in the OB release 2 prior to the Super Carrier and with A we've no idea what is actually being done server side etc if anything or client side to drive all that.. Though we saw all 3 items drop at once so it's hard to work out what caused the impact.

 

Just my 2$ to the discussion.

 

 

 

 

I know this is a little bit off-topic: But can you tell me what I would miss in the stable version from the beta one? Any significant modules that wouldn't function anymore (besides The Channel map which if I recall correctly is only in open beta right now)?

 

Would love to be able to play with the F/A-18, Tomcat, and some WW2 warbirds (p-47 is beta too, right?).

 

 

Cheers :joystick:

[squirrelmeister]

I know this is a little bit off-topic: But can you tell me what I would miss in the stable version from the beta one? Any significant modules that wouldn't function anymore (besides The Channel map which if I recall correctly is only in open beta right now)?

 

Would love to be able to play with the F/A-18, Tomcat, and some WW2 warbirds (p-47 is beta too, right?).

 

 

Cheers :joystick:

 

edit: will test the stable version and will report back!

 

 

 

 

Nevermind. I am downloading the stable version right now and using the non-steam version. With this I can have both versions and can see for myself what's eventually missing and what not.

 

 

 

 

Cheers :joystick:

Edited June 18, 2020 by squirrelmeister
adding stuff

[Eldur]

So I bit the bullet on TGW's server and rolled us all back to stable and we all instantly noticed almost a 20 - 40% increase in performance on our machines measurable in both Monitor and VR, but not only that there is a significant increase in the server side reported FPS if your using something like Tacview to monitor said performance.

 

Here's just some things we noticed performance wise over the last 4 weeks or so... Prior to the SC dropping we saw server frames of about 70-90fps base line dropping over time down to about 60fps with a lower limit of 40 on our 'heaviest' mission at the time.

 

I was under the impression that there is a dedicated server already that runs without graphics. Does it still have that non-working AI issue?

 

As for the drop itself, you mentioned the reason: Tacview. Or somthing related to that. Disabling it nearly doubled my framerate with the latest versions that had them drop down to the drink.

[speed-of-heat]

edit: will test the stable version and will report back!

 

 

 

 

Nevermind. I am downloading the stable version right now and using the non-steam version. With this I can have both versions and can see for myself what's eventually missing and what not.

 

 

 

 

Cheers :joystick:

 

Ground Radar? P-47 if you have it ?

[squirrelmeister]

Ground Radar? P-47 if you have it ?

 

 

Ground Radar too? So no green screen for me :megalol:

 

The p-47 is only on the beta branch that I do know.

 

 

 

Two things I can live without if the performance is better. But everything under 10 FPS more and I am back in beta ^^

 

 

 

 

Cheers and thanks man! :joystick:

 

 

 

 

edit: oh yeah, still downloading

Edited June 18, 2020 by squirrelmeister
grammar

[Vertigo72]

I was under the impression that there is a dedicated server already that runs without graphics.

 

Not running graphics does not really make it a "dedicated server". Its still like 98% of the game. Or more like 100% just with a null render device.It actually has frames per second.

 

For perspective, I fly another niche flightsim, condor, a soaring simulator. As different as it is, it actually has much in common with DCS, like niche market, small dev team like laser focus on realism, flight model , physics.. But one very important difference is that it has a dedicated server app. It uses 6Mb ram. Mega, not giga. I checked, it literally uses less ram than windows calculator. And has no problem supporting 64 players lag free with literally unmeasurable cpu load. THe only restriction is bandwidth, if you have that, running 2 dozen servers on any old obsolete 10 year old pc is totally no problem.

 

Of course, its much simpler to achieve than in a combat sim. There are no bullets to keep track off. No rockets, no radar, no sams, no AI (well none to speak off, the tow planes are AI I guess). The 'damage model' is a little less elaborate and not much more than counting the number of wings and control surfaces that are attached to the fuselage. But it also totally doesnt care about anything other than syncing the players world, it doesnt care about your flight model or damage model or the atmospheric model or anything of that. It doesnt calculate collisions, physics, it doesnt do the atmospheric modelling of thermals or slope winds, it doesnt know where those clouds are or how strong the thermal underneath it, let alone that it would do any sort of rendering, only the clients do. But they all get the same result, because the physics model is deterministic. For instance each client gets a seed value for the weather and all clients wil calculate the exact same weather based on that seed (and map and weather settings and time of day etc etc), there is no need for the server to even care, let alone do those same calculations. What good would it do even if it came up with a different result ?

 

I dont know how that works in DCS. Like, who gets to decide if a bullet hit or not. Presumably its the plane shooting who "owns the bullets" and determines their flight path and tells everyone else about those bullets, and probably any plane that gets hit by those bullets that where said to follow that fly path determines if they got hit and what damage it causes, and then tell everyone else. Im guessing, but it seems logical. What the role of the server in this? It should be none. Just making sure both planes are aware of those bullets and their effect. but calculating the ballistics and collision detection, why redo that? Each client already does those calculations and it works in single player or "peer to peer", duplicating all that work on a "dedicated" server seems extremely wasteful to me. You need to check the client isnt running some mods that shoot magic bullets or have invulnerable planes (both sims do) but then if one clients says "my bullets went there" and another "I just got hit by those bullets, took me left engine out", just take them on their word, and tell everyone else. Why check his math by redoing it, using the same data and the same algorithms?

 

Maybe its needed for AI though. I suppose someone has to decide what that sam battery or AI plane does. Although even that may not be needed when their behavior is completely deterministic. Im actually not sure how that works in condor, "who flies" the AI tow planes. Maybe its just the client that is being towed, and then tells the server the position/attitude etc of the tow plane, and the server than broadcasts that to everyone else. That would make it doable for condor to ignore the AI in the dedicated server, but that wouldnt really work in DCS. Which clients gets to control the behavior of that sam battery or mig 29? But Ai is about the only thing I can think off that a DCS server may need to calculate. AFAICT it should be able to ignore everything else and leave it up to the clients, and just ensure their worlds are in sync. Given the amount of "stuff" that needs to be synced, I can see DCS being much more bandwidth hungry that something like Condor, but it shouldnt take 32GB and a monster PC to keep client worlds in sync and perhaps do the AI.

Edited June 19, 2020 by Vertigo72

[squirrelmeister]

Hello there!

 

 

So now I can say that, yes, stable version gives me deffinetely more performance. And I have to say I am getting like 15 up to 40! mor FPS. So with that said, I am gonna stay on stable.

 

 

But I have a problem with the shader mod: it won't work, not the kegetys version and neither your one speed-of-heat. The following compiling error comes up no matter what version I tried. And yeah, I always do a repair after it is not working.

 

 

Anybody can help? (I mean I am happy with the performance in stable - but even more FPS? WHY THE HECK NOT!)

 

 

Cheers :joystick:

[speed-of-heat]

you need to use the earlier version of the shader mod for stable ... would be my guess

[squirrelmeister]

you need to use the earlier version of the shader mod for stable ... would be my guess

 

 

Yeah, that was my mistake!

 

 

And here are my personal conclusions (yet again :lol:) from testing today in the stable version (non-Steam and in VR using Rift S):

 

 

 

 

Way better performance overall. 15-40 FPS more than compared to beta+shader mod! And I tested kegetys version for 2.5.6.49314 (latest compatible version for up to date DCS stable version) which gave me not really that much of a boost - actually I couldn't "really" see a difference in FPS (there is some but maybe max 5 FPS more).

 

 

In that case I decided to go without the shaders mod (looks better too as we all know). And fortunately I could theoretically push visible range up to high because the FPS won't go under 40 even with MSAA x4. And what I saw too what always wondered me in beta: My FPS changed if I changed the MSAA, I never see that in beta. And to put one more thing on this: in beta it wasn't possible for me to go higher than a PD of 1.0 even with MSAA off completely which I always felt was strange, even when I lowered the PD to like 0.5 I couldn't get a single FPS more out of it, that's why I thought this all comes down to the CPU which also was strange since I don't think that my 7700k @ 4.9 MHz and my 1080ti could be that bad- but in stable I have to say DCS feels like a completely different game performance wise. I see differences in FPS when I am changing settings which, like I said before, strangely wasn't the case in beta.

 

 

 

So the big conclusion (how many do I have now? haha :lol:):

Something is wrong with the beta version at least in my case.

And stable is the way to go when there are performance problems with the beta version.

 

 

I would love to see others testing this and maybe you see the difference in FPS too.

 

 

 

The attached screenshots are showing my ingame settings, the nvidia controll settings and an inflight screenshot from the instant action free flight mission in PG with the F/A-18. (same thing with the performance in Caucasus compared to beta!)

 

 

And with that I am now going to have a nice cold Heineken!

 

 

Cheers :joystick:

[speed-of-heat]

i did some testing in pancake mode, and there was between 20 and 40 fps difference which is about 20-40% ... bear in mind that if you have ASW turned on you wont see much difference in FPS unless it drops below half your "screen rate" e.g. 40 on the Rift... also if you are using IC Pass variant of the mod the perf benefit is pretty marginal

Edited June 19, 2020 by speed-of-heat

[squirrelmeister]

i did some testing in pancake mode, and there was between 20 and 40 fps difference which is about 20-40% ... bear in mind that if you have ASW turned on you wont see much difference in FPS unless it drops below half your "screen rate" e.g. 40 on the Rift

 

 

 

 

Yeah, I know that it not really makes a difference when using ASW. But as I said stable feels also better than beta in general and for me I think it is better to have more places where it runs at full 80 FPS which is the case in stable compared to beta.

 

 

 

 

Cheers :joystick:

[squirrelmeister]

Yeah, I know that it not really makes a difference when using ASW. But as I said stable feels also better than beta in general and for me I think it is better to have more places where it runs at full 80 FPS which is the case in stable compared to beta.

 

 

 

 

Cheers :joystick:

 

 

 

 

But and I just thought of that: My FPS in beta go under 40 FPS for around 75-90% of the time with the same settings as compared to stable where I just don't see FPS under 40.

 

 

So for me at least I will stay on stable for the time being. Although I miss the Channel Map and I would love to see this map in stable and how it performes.

 

 

 

Cheers :joystick:

Edited June 19, 2020 by squirrelmeister
adding

[squirrelmeister]

i did some testing in pancake mode, and there was between 20 and 40 fps difference which is about 20-40% ... bear in mind that if you have ASW turned on you wont see much difference in FPS unless it drops below half your "screen rate" e.g. 40 on the Rift... also if you are using IC Pass variant of the mod the perf benefit is pretty marginal

 

 

 

 

Yeah, your mod gave me a boost in FPS in beta but I still see more FPS in the stable version without any shader mod compared to the beat with the mod.

 

 

 

 

Cheers :joystick:

[speed-of-heat]

yep

[Gryzor]

I noticed in latest 2.5.6.50979 open beta a parameter which hurts fps near 10% or so: concretely, in file terrainoptions41.lua.

 

In that file, there is a line which saids:

multiAppendBufferSize = 524288;

 

If you put:

multiAppendBufferSize = 400000; you will notice a 10% increase in FPS (at least, it seems to me an increase). In fact, I look old beta&stable versions and the parameter by default were 400000. It seems that new parameter of 524288 cause some fps issues.

 

Devs seriously, look more in detail about optimization, this is a mess for users every time editing files.

Edited June 19, 2020 by Gryzor

[speed-of-heat]

nice find!

 

on an inital check that seems to work for me as well!

[speed-of-heat]

also changing it seems to be IC compliant

[petsild]

In my 2D view I can't tell the difference, exactly what the parameter is doing?

[diogofalcao]

I forced physx to cpu (nvidia control panel) and now I get less stutters, not sure if it's just my system but it's so smooth now (could be placebo effect and I don't even know if DCS uses physx), but someone else could try and tell us.

[Vertigo72]

DCS doesnt support physx, so I cant see how it matters.

 

I fear this is a common thread, DCS performance is difficult to measure in a way that is controlled and reproducible, and even when you try hard to control the variables, it just seems varies. Which leads people to believe / promote all kinds of snake oil performance boosting tricks.

[Charly314159]

I noticed in latest 2.5.6.50979 open beta a parameter which hurts fps near 10% or so: concretely, in file terrainoptions41.lua.

 

In that file, there is a line which saids:

multiAppendBufferSize = 524288;

 

If you put:

multiAppendBufferSize = 400000; you will notice a 10% increase in FPS (at least, it seems to me an increase). In fact, I look old beta&stable versions and the parameter by default were 400000. It seems that new parameter of 524288 cause some fps issues.

 

Devs seriously, look more in detail about optimization, this is a mess for users every time editing files.

 

It is really troublesome to see here again to see that it seems ED has little to no control on DCS evolutions and possible side effects.

[Vertigo72]

It is really troublesome to see here again to see that it seems ED has little to no control on DCS evolutions and possible side effects.

 

Alternatively, its possible ED changed that parameter for a good reason that we can only guess at unless they tell us. Admittedly, if it really does make a 10% difference, that had better be one good reason, but I dont notice any performance impact.

[Charly314159]

Alternatively, its possible ED changed that parameter for a good reason that we can only guess at unless they tell us. Admittedly, if it really does make a 10% difference, that had better be one good reason, but I dont notice any performance impact.

 

Sure maybe. But considering the current level of performance I don't see anything justifying a 10% drop

[SignorMagnifico]

I forced physx to cpu (nvidia control panel) and now I get less stutters, not sure if it's just my system but it's so smooth now (could be placebo effect and I don't even know if DCS uses physx), but someone else could try and tell us.

 

Genius! I tried this solution, and it worked for me. I was getting about 27 - 30 FPS on the supercarrier on Mission 3 of Cage the Bear for the F-14. After forcing PhysX to CPU, I get about 31 - 35 FPS. Another side benefit is that this solves the MiG-21 radar FPS bug. There is a bug with the latest nVidia drivers that causes my FPS to drop by 50% (from 60 to about 30 or less) when the radar is turned on and pointed at the ground. However, with PhysX set to CPU only, I don't experience the FPS drop and can maintain about 80 FPS with the radar on with the ground. This guy is on to something!

Edited June 23, 2020 by SignorMagnifico

[hreich]

Genius! I tried this solution, and it worked for me. I was getting about 27 - 30 FPS on the supercarrier on Mission 3 of Cage the Bear for the F-14. After forcing PhysX to CPU, I get about 31 - 35 FPS. Another side benefit is that this solves the MiG-21 radar FPS bug. There is a bug with the latest nVidia drivers that causes my FPS to drop by 50% (from 60 to about 30 or less) when the radar is turned on and pointed at the ground. However, with PhysX set to CPU only, I don't experience the FPS drop and can maintain about 80 FPS with the radar on with the ground. This guy is on to something!

 

How do you force Physx to cpu only?