game engine Game Engine improvements discussion

[Stoikiy]

As described in the developer diaries, a significant amount of work has been done to update the engine compared to the previous titles in the Great Battles series.

I would like to know whether any conceptual changes have been made to the AI and the netcode. As far as I remember, in the original IL-2 from 2001 the AI controlled the aircraft as a whole. This resulted in less precise simulation, but it required far fewer resources, allowing us to see large aerial formations.

In IL-2 Great Battles, the AI behaves like a real pilot, moving the stick and throttle in the same way a player would. In theory, this sounds impressive, but in practice it led to situations where only about 32 aircraft could be present in a small area before the CPU load became too high.

Has any work been done in this direction, and will it be possible to see 100–150 aircraft in the air simultaneously without major performance loss in both single-player and multiplayer?

It is also interesting to know whether modern upscaling and anti-aliasing technologies such as DLSS, DLAA, FXAA, XeSS, and Frame Generation will be implemented.

[Dash,Polder]

I truly hope they remove the AI flaps wanking in every fight, this is not the norm or the technique would be published in every pilots handbook and operators manual.  

That's just plain nuts and gaming the game.

[LukeFF]

Yes, we've talked about AI improvements in our dev blog videos. it will be a lot different than GB.

[Dora]

Will there be improvements for RDNA 3 GPU users?

 It gets HALF the performance of same generation Nvidia cards on Il-2 Great Battles , AFAIK the only game with that problem.

[airsheep_VR]

As a VR player I would like to know if DLSS and fixed foveated rendering will be implemented? Old Il2 games crashes with FFR enabled but it is necessary to reduce the gpu workload. DLSS also a good way to improve visuals as it works well with the other flight sims.

[LukeFF]

1 minute ago, airsheep_VR said:

As a VR player I would like to know if DLSS and fixed foveated rendering will be implemented? Old Il2 games crashes with FFR enabled but it is necessary to reduce the gpu workload. DLSS also a good way to improve visuals as it works well with the other flight sims.

Please see Han's comments here: https://www.reddit.com/r/IL2Series/comments/1oqpwh0/comment/nnnpbe5/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button

[Dash,Polder]

39 minutes ago, airsheep_VR said:

As a VR player I would like to know if DLSS and fixed foveated rendering will be implemented? Old Il2 games crashes with FFR enabled but it is necessary to reduce the gpu workload. DLSS also a good way to improve visuals as it works well with the other flight sims.

Better yet support actual eye tracking and the latest features of cutting edge HMD's, might just grab one than for another upgrade.

[Col.Sanders]

6 hours ago, LukeFF said:

Please see Han's comments here: https://www.reddit.com/r/IL2Series/comments/1oqpwh0/comment/nnnpbe5/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button

It's interesting that Han basically says it doesn't help but provides no information on how it was tested. Dynamic foveated rendering should be a industry norm going forward to help with high resolution headsets. When testing VR what headsets did you all use? The concern many have is that DFR is being written off for a game engine that should be lasting another decade. Seems like a huge miss. 

[airsheep_VR]

10 hours ago, LukeFF said:

Please see Han's comments here: https://www.reddit.com/r/IL2Series/comments/1oqpwh0/comment/nnnpbe5/?utm_source=share&utm_medium=web3x&utm_name=web3xcss&utm_term=1&utm_content=share_button

Thanks, I understand his points but still thinks if that feature can help in MSFS & DCS a lot than still be helpful here too. Maybe it is less than 10% advantage but vr players fight for every FPS 🙂
Also, do you have any news about DLSS? I haven't found any info...In MSFS24 I can see how good it is, the picture is clear and the gpu & cpu load is way better. 

Edited November 29, 2025 by airsheep_VR

[LukeFF]

6 hours ago, Col.Sanders said:

When testing VR what headsets did you all use?

That would be a question for Han to answer in terms of specifics, but in general, they have the ones that are in common usage, that much I know.

2 hours ago, airsheep_VR said:

Also, do you have any news about DLSS? I haven't found any info...In MSFS24 I can see how good it is, the picture is clear and the gpu & cpu load is way better. 

The plan is to have FSR at launch but not DLSS. 

[Aapje]

1 hour ago, LukeFF said:

The plan is to have FSR at launch but not DLSS. 

And then the plan is that I will whine at you guys until you add that too 😛

[Avimimus]

18 hours ago, Stoikiy said:

As described in the developer diaries, a significant amount of work has been done to update the engine compared to the previous titles in the Great Battles series.

I would like to know whether any conceptual changes have been made to the AI and the netcode. As far as I remember, in the original IL-2 from 2001 the AI controlled the aircraft as a whole. This resulted in less precise simulation, but it required far fewer resources, allowing us to see large aerial formations.

In IL-2 Great Battles, the AI behaves like a real pilot, moving the stick and throttle in the same way a player would. In theory, this sounds impressive, but in practice it led to situations where only about 32 aircraft could be present in a small area before the CPU load became too high.

Has any work been done in this direction, and will it be possible to see 100–150 aircraft in the air simultaneously without major performance loss in both single-player and multiplayer?

It is also interesting to know whether modern upscaling and anti-aliasing technologies such as DLSS, DLAA, FXAA, XeSS, and Frame Generation will be implemented.

 

There was a lot of pressure (form the Il-2 1946 era) from people who wanted the AI to use exactly the same flight model and physics model as the player aircraft. This is obviously very computationally intensive. There is work being done to allow bombers (and perhaps even some fighters) to use simplified systems models to allow larger formations, and this is an announced feature in Korea.

However, for something like multiplayer, my understanding is that calculations need to be done centrally (i.e. on the host/server) to make sure there isn't desynchronisation, so it will be very computationally intensive, as the host has to run the full flight models for all of the players - at least that is my understanding. I think this may mean there is a bottleneck that exists separately from the netcode... and it might be a fundamental tradeoff between the number of players and the quality of flight/phyics/damage models. However, I could be wrong about this... there could be a clever work around that I don't know about.

[Stoikiy]

48 минут назад, Avimimus сказал:

from people who wanted the AI to use exactly the same flight model and physics model as the player aircraft.

Sometimes people don’t know what they really want, or they don’t understand what problems it may lead to. 
If I have to choose between visual spectacle and an AI pilot that needs to be trained to fly at least approximately at a human level, I would choose the first option.

 

52 минуты назад, Avimimus сказал:

However, for something like multiplayer, my understanding is that calculations need to be done centrally

The closest competitors are able to place more than 300–500 ground units in online matches without any significant performance issues. 
You can also build these ground units yourself, creating incredible defensive lines. I even have a meme about this:

 

Скрытый текст

 

[Avimimus]

47 minutes ago, Stoikiy said:

The closest competitors are able to place more than 300–500 ground units in online matches without any significant performance issues. 
You can also build these ground units yourself, creating incredible defensive lines. I even have a meme about this:

Ground units have very simple AI, very simple damage models, almost no systems models.

I was specifically referring to objects with complex flight models, damage models, and systems models (such as flyable aircraft). One couldn't do that with aircraft unless one was happy with much lower fidelity (probably closer to that you'd get back in the 1990s).

I had understood we were talking about the number of human flown aircraft in a single battle.

[AEthelraedUnraed]

3 hours ago, Stoikiy said:

Sometimes people don’t know what they really want, or they don’t understand what problems it may lead to. 
If I have to choose between visual spectacle and an AI pilot that needs to be trained to fly at least approximately at a human level, I would choose the first option.

I for one never liked the old 1946 AI/physics model, that would do some entirely impossible manoeuvres.

Both options have their advantages, and these are not equally large in every situation. For fighter aircraft, I like it when the AI has to deal with the exact same physics as I, since you're really flying the plane to its limit. For bombers it doesn't matter as much since these don't manoeuvre much anyhow. Which is the reason they're developing a simplified flight model for the B-29, so that we'll be able to have a large quantity of them in the air.

[Reisen]

I’d rather the AI ‘cheat’ a little if they can at least put up an interesting fight. The Great Battles “if enemy=true, turn circles for infinity” AI routine gets really boring. Hopefully we see something much better in Korea. It will probably be one of the biggest deciding factors for me since I haven’t preordered.

[AEthelraedUnraed]

6 minutes ago, Reisen said:

I’d rather the AI ‘cheat’ a little if they can at least put up an interesting fight. The Great Battles “if enemy=true, turn circles for infinity” AI routine gets really boring. Hopefully we see something much better in Korea. It will probably be one of the biggest deciding factors for me since I haven’t preordered.

AI turning circles has next to nothing to do with whether or not it uses the same flight model as the player does.

[Kubert]

I don't see turning in circles as a problem. It mostly happens in 1v1 scenarios which is not that common in campaign or career. Just having another enemy nearby can be enough of a challenge against the AI. Especially when flying a worse plane. I'm not even talking about scenarios like 4v6  or 2v5 and so on. For the AI to be "good", it should be made specific for each plane. So that the P-47 never turns, but the Spitfire does. And a balanced plane like the BF-109 would have to be able to assess the situation, so that it turns against the La-5 but not against the Hurricane. I doubt that anyone will ever program such an AI and I don't want to go back to the cheaters from the IL-2 1946 who flew like a UFO.

1v1 against a turning enemy can also be fun to fly. Especially when enemy have a more maneuverable aircraft than me. Then instead of turning behind him, I pick up speed and try to get into vertical maneuvers. I've turned with them in the past and it was frustrating way to fly. Following isn't always the way...sometimes is needed to figure out something else.

[FuriousMeow]

12 hours ago, Avimimus said:

 

However, for something like multiplayer, my understanding is that calculations need to be done centrally (i.e. on the host/server) to make sure there isn't desynchronisation, so it will be very computationally intensive, as the host has to run the full flight models for all of the players - at least that is my understanding. I think this may mean there is a bottleneck that exists separately from the netcode... and it might be a fundamental tradeoff between the number of players and the quality of flight/phyics/damage models. However, I could be wrong about this... there could be a clever work around that I don't know about.

All players calculate their FM locally. The host only calculates the AI FM, but also handles data transfers to communicate to all players where each person is and the position of AI and what they are doing. Additionally, there's the communication of evety aspect of what each player is doing. Most games the clients communicate the planes speed, vector and basic states - gear up/ down and flaps position, firing/ launching weapons, etc. But with RoF, and thus GB and almost certainly Korea, the clients inputs are also transmitted so the various control surface deflection and some other more fine details are communicated which adds to the data being transferred and computed. If all players FMs were calculated on the server, it would be a laggy mess all around. Clients always calculate locally what they have control of, which is just the client's plane, what its doing and what damage it does to other clients or AI, and what its colliding with. 

For Aces High, for example, in order to have so many players on a single server the data transmitted for the flying state is very little and set at a low tick but there's a computation in the background of the clients and server that extraoplates flying vector based on previous states so the flying path seems very smooth but is using far less data than you'd think. Which is why there can be some interesting behaviors from other players in poor connection scenarios.

I'm sure if there was a dedicated net code specialist, and revising all the data that's transmitted, the player count could be increased without a negative impact. 

Edited November 29, 2025 by FuriousMeow

[Stoikiy]

9 hours ago, Reisen said:

I’d rather the AI ‘cheat’ a little if they can at least put up an interesting fight.

one hundred percent my opinion. I'm with you here.  I remeber that fights in BoS and BoM was very boring.  This is why i'm always played PvP

[[I.B.]-=ViRUS=-]

5 часов назад, Stoikiy сказал:

one hundred percent my opinion. I'm with you here.

мало ты играл в старичка, где по рельсам боты летали. Когда эмиль в пологом климбе от Як-1б уходил - веселья было еще меньше. Уж лучше так как есть.

[Shinobimono]

19 минут назад, [I.B.]-=ViRUS=- сказал:

Уж лучше так как есть.

Can't agree with this. Great Battles' AI behaves like a rookie: predictable, boring. Making very stupid wobbling in turns (common for FW-190A in fights with P-51).

6 часов назад, Stoikiy сказал:

one hundred percent my opinion.

IIRC devs will not make same mistake twice, so we will have simplified FM for large formations, as was mentioned above, and simplified FM for remote AI aircraft.

 

[[I.B.]-=ViRUS=-]

2 часа назад, Shinobimono сказал:

Can't agree with this. Great Battles' AI behaves like a rookie: predictable, boring.

В старичке не сильно лучше. Мало летали в него. Когда у оппонента ниоткуда берется пара лишних сот лошадей, движок не кипит, самолёт не срывается в штопор на висении - нафиг эти "рельсы"

[Shinobimono]

1 час назад, [I.B.]-=ViRUS=- сказал:

Когда у оппонента ниоткуда берется пара лишних сот лошадей

Зато как они там маневрировали в бою, не чета местным. 

Плюс всё это поведение в Забытых Сражениях зависит от версий игры, коих набралось не мало.

Если Забытые сражения я уже помню поверхносто, конечно, то здесь одиночной игры я наелся достаточно, чтобы сказать, что такие боты не вызывают вообще никакого интереса в игре. Атмосферы нет. 🤷🏼‍♂️

[Trooper117]

If you are going to converse in this forum in a foreign language, can you at least put a translation in with it as well... thank you!