Game engine performance with high number of airplanes.

[=Black=Cocaine]

Hey there, 

I'm trying to play the Battle of Britain campaign with high number of airplanes in VR, but as soon as the enemy airplanes spawn in the game goes into slow motion.

Is this normal, I mean is this expected? Can anyone run high number of airplanes?
I'm running the game on R9 5950x and Asus TUF 4090.

according to windows il2 doesn't even stress the system down too much (about 10% on the CPU ) 

am I missing something and being a total idiot here? 

[PatrickAWlson]

Running the game with a large number of planes, especially bombers, will bring it to its knees.

[=Black=Cocaine]

Thank you for your answer.
Is that due to  game engine optimization ? I've seen the argument of how CPU consuming is to have many units, but is it really ? Other simulators with high number of units and modern graphics run just fine on my specs.

[PatrickAWlson]

14 hours ago, =Black=Cocaine said:

Thank you for your answer.
Is that due to  game engine optimization ? I've seen the argument of how CPU consuming is to have many units, but is it really ? Other simulators with high number of units and modern graphics run just fine on my specs.

 

It does seem to be CPU bound but I have never done anything beyond observing cause and effect - i.e. I have no numbers or patterns.  The community has always had a WHY??? attitude towards this game's performance.  I don't know anything about the games internals that would be needed to state specifics.  To steal a phrase from Game of Thrones: "It is known".  Usually stated with great conviction on a topic about which one knows nothing.

[Avimimus]

I suspect it is related to the fact that the aircraft used by the AI have the same detailed modelling of systems as the player aircraft (and in spite of the lack of clickability, the modelling is actually very deep). One of the recurring complaints about the old Il-2 was the AI aircraft having simplified modelling which occasionally gave them an advantage, this isn't the case anymore. In contrast, DCS simplifies the flight models and systems modelling for AI aircraft, while Cliffs of Dover is an older game at this point (it was unrunnable on most machines when it was released though).

 

However, this is mainly speculation on my part.

[1PL-Husar-1Esk]

13 hours ago, Avimimus said:

I suspect it is related to the fact that the aircraft used by the AI have the same detailed modelling of systems as the player aircraft (and in spite of the lack of clickability, the modelling is actually very deep). One of the recurring complaints about the old Il-2 was the AI aircraft having simplified modelling which occasionally gave them an advantage, this isn't the case anymore. In contrast, DCS simplifies the flight models and systems modelling for AI aircraft, while Cliffs of Dover is an older game at this point (it was unrunnable on most machines when it was released though).

 

However, this is mainly speculation on my part.

Would prefer simplyfied ai because current might have same simulation as player plane but AI is dump and easy to kill , so no benefit for challenging dofight just system hog.

[=Black=Cocaine]

So what you guys are saying is that no one every was able to run the PW dynamic campaigns with high number of airplanes due to it overloads the game's engine ? 
So at this point the bottle neck is the engine not the hardware....

Btw I've managed to find a combination of settings in PWC to make it playable on my spec and with medium number of planes.
I've turned of the smoke altogether on the map, set AA units to low, ground units to minimum, cpu consuming units to low, structures to low and it runs smoothly for most of the time I have to say. Not always tough. I think I will play around with the number of bombers as PatrickAWlson have mentioned they are power consuming.

 

 

[JG27*Kornezov]

I think the best thing is to make dynamic scaling.

 

The basic idea is only AI planes that are within a determined radius of a human player to have a full model. e.g. 2 km. or 5 km, or  even 1 km. For the rest to use a simplified model that models correctly speed and climb and turn rate.

 

It is a waste of resources to run the same complex flight model for all AIs. I think this is an opportunity for the developers, because it is the only way to make  large scale single player (even multiplayer with AIs), experience. Historically some of the fights over Europe included more than 100 planes. Just imagine to see a bomber formation of 100 planes.

For me the best combo is realistic clouds and big bomber formations. I think things  to the ground has reached the limits of current possibilities (except if we look from tanker's perspective but this is a whole different story).

Edited November 7, 2023 by JG27*Kornezov

[1PL-Husar-1Esk]

1 hour ago, JG27*Kornezov said:

The basic idea is only AI planes that are within a determined radius of a human player to have a full model. e.g. 2 km. or 5 km, or  even 1 km. For the rest to use a simplified model that models correctly speed and climb and turn rate.

Not really a solution, when you are close to that 100 bombers you about to engage (most important time) and complex simulation kicks in , yours performance will drop significantly. 

[Avimimus]

That is one of the issues with the American strategic bombing campaign... dense box formations - a fighter could be simultaneously within range of the gunners of 36 B-17 bombers (sometimes more).

[JG27*Kornezov]

Just imagine a box. How many bombers you can fit in without the system starting to struggle. It is a completely unrealistic to say that in a box of 1 km you will fit 100 bombers in a tactical scenario. The idea is that you do not need to precisely  compute interactions of stuff that is 10 km. away.

 

The other simpler solution is to add b17 and b24 and only for those models to make simplified flight model. You are not going to make aerobatics for them so a simple behavior in critical flying regimes will be completely enough.

Edited November 10, 2023 by JG27*Kornezov

[=Black=Cocaine]

I'm sorry guys I might be missing something but why is everyone keep saying that this is a system issue? 

The CPU clocks on 8% whilst running the game in VR, the 4090 is somewhere 30% if I remember correctly. I don't think this is a system issue at all. I also don't think calculating that much simulation even with a 100 planes would be incredibly CPU killer these days.

My computer play back high end cinema grade uncompressed camera raw footages without a hick up. And that actually puts the system under proper stress.

 

[JG27*Kornezov]

This is just a theory we do not know as it is a black box. But in science they do grids when trying to emulate complex differential equations over a space, as it is not feasible to compute everything.

Edited November 10, 2023 by JG27*Kornezov

[=Black=Cocaine]

11 minutes ago, JG27*Kornezov said:

This is just a theory we do not know as it is a black box. But in science they do grids when trying to emulate complex differential equations over a space, as it is not feasible to compute everything.

Ah okay

I never heard about this before but I get what you mean. Interesting 

[PatrickAWlson]

2 hours ago, =Black=Cocaine said:

I'm sorry guys I might be missing something but why is everyone keep saying that this is a system issue? 

The CPU clocks on 8% whilst running the game in VR, the 4090 is somewhere 30% if I remember correctly. I don't think this is a system issue at all. I also don't think calculating that much simulation even with a 100 planes would be incredibly CPU killer these days.

My computer play back high end cinema grade uncompressed camera raw footages without a hick up. And that actually puts the system under proper stress.

 

 

A possible explanation is a tangled wait state for the different threads.  That would slow the game down horribly while actually reducing CPU.  As Kornezov said, without being part of the team and having the ability to profile the code you're not going to know.  The one thing that I have found is almost 100% true is that the source of poor performance is not what you thought it was going to be before you actually profile.  That is especially true for situations like this that have lingered for 14 years.  If it was easy they would have solved it by now.