Source of performance problems

I have a pretty beefy machine and can run DCS at insanely high settings in VR, i can hold stable 45 fps without problems in OB 2.5.5 or 2.5.6 ...

The only reason for heavy performance drops on my end are AI objects in vicinity, it's not the new shaders or eye candy. Modern GPU's (1080 - 2080ti) eat pixelshaders for breakfast, so it's definitely not the "new lighting".

Please ED, focus all of your optimization at this particular matter! It's been there for years and slashes back everytime! There is no mission or campaign i can play in this state. In 2.5.6 my fps drop from 45 to 20 in the same scenario.

Check those screenshots (made in 2.5.5 latest OB)

For testing I used Maple Flag's Aggressor Campaign mission "Arrival", which is a mission where you watch a transport coming in to land at Nellis. First two screenshots are 2.5.5, the next two are 2.5.6. The fps are around 15 or less throughout in 2.5.6. I get 80 at first in 2.5.5 and then 40 (ASW) as I get closer to the runway. The graphs from MSI AB on the left show the stats from the start of the mission until the transport lands in each case.

Notice that at the initial menu screen there is an fps drop of around 20%. This is probably in the ballpark (15%) of what I am seeing in missions where there are no other objects and the severe frame rate drop does not occur. Although not a disastrous frame rate drop, in VR we really need every bit of performance, so this is most unwelcome. From my past experience, changes in fps at the main menu do reflect fps changes in the game. Are you also seeing this?

I notice that you are seeing your CPU max out at 100%. Curiously I am not seeing that, so I'm puzzled as to what's going on: CPU at around 70%, and GPU at 13%. I am entirely unfamiliar with your HMD and its details.

I have ASW on and as soon as 80 fps can't be maintained it drops to 40%, hence the sudden cliff-edge drops in 2.5.5.

Settings are the same for both tests.

The white squares are my labels, which should be almost transparent, but that was broken in an update over a year ago, and has never been fixed. Just saying ED. Very annoying as this is vital, imho, in VR.

I am also enhancing AA by using 2x w/ MFAA in nVidia Control Panel.

I have a pretty beefy machine and can run DCS at insanely high settings in VR, i can hold stable 45 fps without problems in OB 2.5.5 or 2.5.6 ...

The only reason for heavy performance drops on my end are AI objects in vicinity, it's not the new shaders or eye candy. Modern GPU's (1080 - 2080ti) eat pixelshaders for breakfast, so it's definitely not the "new lighting".

Please ED, focus all of your optimization at this particular matter! It's been there for years and slashes back everytime! There is no mission or campaign i can play in this state. In 2.5.6 my fps drop from 45 to 20 in the same scenario.

 

Check those screenshots (made in 2.5.5 latest OB)

I think you have a point here! It is also my personal feeling that stutters occur when AI aircrafts are active and near.

One can only speculate the exact cause of the performance drop, but everything suggests CPU bound activity. It could be that GPU has to wait AI control logic to finish its calculations if it is done sequentially. Maybe scripting also makes GPU wait causing stutters. Either way the solution might require architetural level changes and optimizations to parallelize these tasks.

For testing I used Maple Flag's Aggressor Campaign mission "Arrival", which is a mission where you watch a transport coming in to land at Nellis. First two screenshots are 2.5.5, the next two are 2.5.6. The fps are around 15 or less throughout in 2.5.6. I get 80 at first in 2.5.5 and then 40 (ASW) as I get closer to the runway. The graphs from MSI AB on the left show the stats from the start of the mission until the transport lands in each case.

 

Notice that at the initial menu screen there is an fps drop of around 20%. This is probably in the ballpark (15%) of what I am seeing in missions where there are no other objects and the severe frame rate drop does not occur. Although not a disastrous frame rate drop, in VR we really need every bit of performance, so this is most unwelcome. From my past experience, changes in fps at the main menu do reflect changes in the game. Are you also seeing this?

 

I notice that you are seeing your CPU max out at 100%. Curiously I am not seeing that, so I'm puzzled as to what's going on: CPU at around 70%, and GPU at 13%. I am entirely unfamiliar with your HMD and its details.

 

I have ASW on and as soon as 80 fps can't be maintained it drops to 40%, hence the sudden cliff-edge drops in 2.5.5.

 

Settings are the same for both tests.

 

The white squares are my labels, which should be almost transparent, but that was broken in an update over a year ago, and has never been fixed. Just saying ED. Very annoying as this is vital, imho, in VR.

 

I am also enhancing AA by using 2x w/ MFAA in nVidia Control Panel.

Did you notice that in game statistics only value that goes up when fps goes down is the "Main: ..., simulation" field. Could this be the culprit of the issue?

It would also be interesting to test this same mission run on server (even on same computer to offload some of the calculations to separate process) and see if there is similar fps drop for the client.

That's why they are working on Vulkan first and multi-threading after.

Without this much needed performance improvement Dynamic Campaign will be impossibile to achieve.

I don't have a particularly beefy machine, but the beta is pretty much unplayable for me even at low settings. I get a nausea-inducing frame rate. I get teleporting aircraft. I get ridiculously long loading times.

The stable release is much smoother at higher settings--Higher settings than the beta, not HIGH settings. DCS is the most resource taxing game I own.

A certain WWII combat sim competitor (which will not be named, ha ha) I can play on high settings with no problems at all, and a frame rate between 90 and 110.

I just tested recreating the same scenario in the mission editor. Just put some groups of AI units on an airflied and visit them after that.

I back up the AI-trashing-FPS theory. I'm currently playing the Snowfox/Clear Field missions. They are rather AI and script-intensive. I have no issues playing VR on Oculus rift S with 40 fps on 2.5.5. However on the newest OB it's basically unplayable as soon as there are more then a couple AI's in the air, instantly going down to 20FPS and staying there.

There is definitely something wrong with AI units, the CPU might be doing extra or doubled but unnecessary calculations. Or it's a deep implemented, but bad concept of code in DCS.

Nothing to do with AI.

https://forums.eagle.ru/showpost.php?p=4214241&postcount=88

(spawn TF-51D Training: Solo takeoff, 1280x768, LOW)

2.5.5 ------22 fps

2.5.6 ------16 fps

 

Lost 6 fps or 27%

 

Yes, my system is a C-64, sue me.

 

Intel Core i7 3630QM @ 2.40GHz (Max Turbo Frequency 3.40 GHz)

16.0GB Dual-Channel DDR3 @ 798MHz

2047MB NVIDIA GeForce GT 635M

447GB KINGSTON SA400S37480G (SATA-2 (SSD))

Nothing to do with AI.

 

https://forums.eagle.ru/showpost.php?p=4214241&postcount=88

Slight fps decrease in 2.5.6 is unsurprising, giving the new lighting and stuff.

In your case the difference is marginal. That's what I'm seeing in most cases when comparing both versions (e.g. 55 fps versus 60).

I tried, however, to add a few AI planes into one of my missions. And got nearly 2x fps loss in 2.5.6 compared to 2.5.5. After sharing my findings on Russian forum, one of the ED developers partially agreed with me that AI might be the culprit here.

So, before saying that this has "nothing to do with AI", add a few AI planes near your spawn point and watch your fps counter go negative.

Nothing to do with AI.

 

https://forums.eagle.ru/showpost.php?p=4214241&postcount=88

I think that there are two distinct effects: a general frame rate loss of up to 20%, as you describe, which can be seen even in missions with no other objects. The other is a much more severe frame rate loss > 80 % when (AI) objects are introduced.

My concern at the moment is that the severe (AI?) object effect is masking the other one, which is being overlooked. A 20% fps decrease is still a serious problem.

Slight fps decrease in 2.5.6 is unsurprising, giving the new lighting and stuff.

20% is NOT "slight" or "marginal". In VR, where we are already pushing limits, it is extremely undesirable.

The question is, is the new lighting "and stuff" worth it? I think, in VR, night is better, but have to question if it has to come at the price of a 20% reduction in fps? By day, I couldn't say.

A 20% fps decrease should not be "unsurprising". It's this sort of attitude that has severely damaged the VR experience over the past couple of years.

I accept that's it's still in beta, and optimisation is probably still ongoing. Let's come back to this when the changes go into stable. In the meantime, anyone who cares about VR should be highlighting any decrease in performance. Once it's gone, it's very difficult to get it back again.

20% is NOT slight. In VR, where we are already pushing limits, it is extremely undesirable.

 

The question is, is the new lighting "and stuff" worth it? I think, in VR, night is better, but have to question if it has to come at the price of a 20% reduction in fps? By day, I couldn't say.

 

A 20% fps decrease should not be "unsurprising". It's this sort of attitude that has severely damaged the VR experience over the past couple of years.

We are not in VR subforum, so I was talking about 2D performance. To me, the new lighting is totally worth it - after the further optimization, of course. Like I said, I don't see a major performance hit when I'm simply flying around - it comes from elsewhere (AI planes in my case).

But I get it that for VR users the every single frame is important.

A 20% fps decrease is unsurprising for the initial release, especially given the very rough state of the 2.5.6 in general. We don't know how much of that decrease came from the new lighting, and how much from "other stuff" or just poor optimization.

Let's come back to this when the changes go into stable.

The whole point of OB and this discussion is to prevent these issues from reaching Stable.

We are not in VR subforum, so I was talking about 2D performance. To me, the new lighting is totally worth it - after the further optimization, of course. Like I said, I don't see a major performance hit when I'm simply flying around - it comes from elsewhere (AI planes in my case).

 

But I get it that for VR users the every single frame is important.

 

A 20% fps decrease is unsurprising for the initial release, especially given the very rough state of the 2.5.6 in general. We don't know how much of that decrease came from the new lighting, and how much from "other stuff" or just poor optimization.

 

The whole point of OB and this discussion is to prevent these issues from reaching Stable.

All fair enough, although the OP was discussing VR. I still haven't gotten over deferred shading, I'm afraid.

My own situation: 2.5.5 OB: 80fps in a mission with the f-18 and nothing else going on. 2.5.6 OB: I have to give up anti-aliasing to keep 80fps. It does make a difference.

The whole point of OB and this discussion is to prevent these issues from reaching Stable.

Sorry, but going completely off topic: in general, how much of what is discussed here is replicated in the Russian forums? Do you get the impression that much of what is discussed here (in English) gets to the developers?

Sorry, but going completely off topic: in general, how much of what is discussed here is replicated in the Russian forums? Do you get the impression that much of what is discussed here (in English) gets to the developers?

The Russian community is much smaller than English, and we tend to discuss module-specific problems, rather than the DCS ecosystem in general. It seems that issues raised on English forums are heavily filtered and only then relayed to the team by the community managers and other lower staff*, while Russian users have a chance to speak directly with the developers.

*In fact, this way is probably more effective.

@-0303-

Holy ... much respect for bringing a C-64 to battle! ;)

Slight fps decrease in 2.5.6 is unsurprising, giving the new lighting and stuff.

 

In your case the difference is marginal. That's what I'm seeing in most cases when comparing both versions (e.g. 55 fps versus 60).

 

I tried, however, to add a few AI planes into one of my missions. And got nearly 2x fps loss in 2.5.6 compared to 2.5.5. After sharing my findings on Russian forum, one of the ED developers partially agreed with me that AI might be the culprit here.

 

So, before saying that this has "nothing to do with AI", add a few AI planes near your spawn point and watch your fps counter go negative.

10% is not marginal and it cannot be linked to "lighting changes" : seriously, you guys have seen the new lighting? It doesn't change anything aside from seeing glowy textures further away. There's nothing that projects shadows or that would reduce performance.

10% is not marginal and it cannot be linked to "lighting changes"

Care to read my further posts after the one you quoted?

seriously, you guys have seen the new lighting? It doesn't change anything aside from seeing glowy textures further away. There's nothing that projects shadows or that would reduce performance.

Have you seen the new lighting? With multiple dynamic lights sources (from the Moon, rockets, street lights etc.) and reworked bloom? Even if you haven't, the increased range of "glowy textures" can't come for free. Else they would've set it to unlimited years ago.

2.5.6 introduced a lot of new things. Many of them are half-baked or not optimized yet. Maybe they'll manage to reduce performance hit from 20% to 10% (or from 10% to 5%). We don't know yet.

What we do know, is that something is causing an enormous performance drop (for some users - well beyond 10% or 20%). And we are trying to figure out what it could be.

i remember there being issues in the past where the TGP on the A10 was rendering even when there was no TGP, might not be related but what ever is going on the work is being doubled up somewhere.

The biggest hits im seeing at the moment is with the contrails and the smoke effects, i have these basically turned off on the slider so there should be nothing other than battlefield smoke and what the jets make. In some instances this is harder to prove as the MP mission running on our server is next to unplayable because of the massive FPS hits being seen, we do have alot of planes flying around in the con layers and i even saw a post here where one person did a bit of a deep dive into how much the jet generated smoke is hitting performance.

As mentioned i don't think the new light is the sole cause, whether it be AI or smoke or unnecessary rendering happening in the background i think ED have got a huge problem as there has been very little communication about what is actually causing the issues, which probably says that they don't know either and are still trying to pin down the root cause themselfs.

RE: Lighting.

TBH, the main thing I noticed is that everything is brighter than it was. Night I think was improved. Outside with lights appearing brighter, and things that should be darker, appearing to be so. But couldn't this just have been done by changing textures and lighting coefficients? Inside the cockpit, the way the landing gear handle (f18c) lights up the cockpit around it and similar effects.

By day nothing in particular shouted out at at me; I did feel that things looked a bit more "like reality", but this could just be down to everything being brighter. Again, TBH, I haven't taken the trouble to compare in any detail. I haven't even bothered turning on SSLR, not even once, after I read about the fps impact of that.

(This is all in VR, to be clear).

Now, in the change log, we get this:

DCS Openbeta 2.5.6.43872 https://www.digitalcombatsimulator.c...a/2.5.6.43872/

 

New cockpit Ka-50. New 3D model, textures, lighting.

As far as I can tell street lights are baked in (Caucasus). TBH I've only tested with moonless nights so far. Wasn't lighting from weapons dynamic already?

The improvement at night was very welcome, but I think (and hope) is only the beginning, as much still needs to be done.

I have sympathy with both your points of view: on the one hand, clearly something fundamental has changed, unfortunately ED seems to not have gone to the trouble of providing details. On the other, practical observable benefits seem to be very few so far, and possibly at great cost. (Sure, still WIP).

As far as I can tell street lights are baked in (Caucasus). TBH I've only tested with moonless nights so far. Wasn't lighting from weapons dynamic already?

I definitely saw a short clip on hoggit, demonstrating cars passing under the dynamic street light and being beautifully illuminated by it. Cannot find it now, sadly.

New (or improved?) weapons lighting was briefly showcased in one of the ED's recent videos (probably "2020 and beyond", but I'm not sure).

Ive just tried to do a mission provided with the F16 module at the nellis range, and with the AI the performance takes a massive tumble.

Seems to be that when any AI aircraft that are also a flyable module have a massive impact on frames in VR.

Not sure if this new lighting is reflecting off of them in a more complex way due to the higher LOD levels? might be an idea to have some sort of "no cockpit version" of the modules that could have slightly less detail to save FPS.

@(504)Blade

i'm currently back on 2.5.5 and face the same problem, but it tanks performance not as hard as 2.5.6. In 2.5.5 fps drop from 45 to 30 which is already extreme, in 2.5.6 it drops from 45 to ~20. When i look at the built-in FPS counter, the Simulation value of the main thread seems to be related to the fps drops. Might be a calculation that is taking too long and slowing down the CPU. My GPU usage shows that there is still headroom for more work, but the CPU is unable to deliver in time.