Performance Megathread

[James Fire]

Eh, ok. So prior to 2.6.x we had maybe one desync every two or three complete multihour games. After 2.6.x as soon as you get desynced u got two choices: 1) drop to menu, rehost the last autosave, wait 3 min for people to download the save and hope desync would hit you 2) save after desync, rehost that save, wait 3 min for people to download the save and get desynced in hour, repeat, desync in 30min, repeat, desync in 15min, 200 years into the game? tought luck, desync every ingame month, literally. My group is not some rare case, plenty people complain on discord and reddit about it. Theres something really wrong with game since 2.6.x and if you have save that desynced once, something is broken and will literally keep desyncing you more and more often into the game. if u know the 'trick' to dont save after desync and roll with autosaves, still you will get desyncs in late game but every 30-60min lets say. Before 2.6.x we literally had maybe one late game desync in last 30hours of playtime, now we get desync every hour, so you were right, humans are prone to error, it may be 30x worse since 2.6.x.

All of group members have same hardware, software, isp (1gb fiber), no problems with connection in other games that we play, it literally changed with patch thats it, no other variables changed, we dont use mods either.

We even had 2 situations when we desynced during save loading when we rehosted the game, we also obv tried different hosts etc.

I was more asking about the lag, as this is the Performance Megathread.

 

[Keldon_]

I was more asking about the lag, as this is the Performance Megathread.

Thing is, the mp lag in stellaris is weird as it comes from problems with sync. Prior to 2.6.x everything was fine, you start the game in mp and normal speed is same like in sp, fastest is same speed like in sp etc. After 2.6.X even fastest is like slow in sp, when u look at calendar you can see day go like this: 1 (3 second break) 23456 (in 0.5 second) 7 (after 5 second) 8910 (in 1 sec) 11 (after 4 seconds) etc. Theres 0 smoothness and you can see when hovering around clock/calendar in mp that other players are lagging 1-3 days behind host, thing is, in mp u can be max 3 days behind host, then host needs to wait for all to catchup which creates that artificial lag. So it not real lag performance wise.

 

[argentAegis]

Except in this case, it is "Works on most machines". You are the outlier, not everyone else. Therefore, something you are doing or have is different.

Yes, which is why I am asking whether anyone else had it and whether anyone had an issue. Instead, you're giving me useless information that it "works on most machines." Great. How does that help me exactly?

 

[ragehavoc]

ryzen 2600
RX 480 8gb
by 2350-2400 1000 star map 24 nations I'm looking at about 10 years every hr.
So to get to end game it would take 15++ hrs, and an entire game on fastest speed would take 25-30+ hrs.
Looking at several twitch videos, it seems this is fairly common, and around that time you also start getting huge hang ups, since all it seems they did was delay certain checks, instead of fixing the way it does it.
So while they may have increased performance, a bit, it is certainly not even close to enough to make the end game playable.

They seriously need to ditch the entire single pop representation and growth system they have because that is the problem and it wont go away until they do.

 

[Su-Pe-Biomech]

It almost starts to freeze in Early Game. I end the game and it works fine again when I restart it...for some time. This is something new.
I don't know, either they need new programmers or they are in dire need of a new, better engine. How it's called, Clausewitz? I think something is wrong with that. It's not even Stellaris alone. There's no way I could play HoI4 til 1945, for example. LAAAG.

 

[Shrike]

I just played a game on 1.0.3 by nostalgia and my god how fast the game used to be ! It was maybe twice as fast as today ! I know there is a lot more mechanics than in the past but I was expecting that all the performance update were somehow useful to counterbalance that.
It is a shame really, but it is also interesting because that means we can easily identify the part of the code that is draining the performance. I know that the pop system is often critized but even in the beginning of the game where there is not a lot of pops the game is not as fast as he use to be. They may be other culprits here (or the pop system is indeed really bad coded !).

Oh yeah, without doubt...but you gotta admit the galaxy is a bit more full now.

After some pretty extensive playing the performance has vastly Improved. So well done developers. If you are reading this devs please continue with further optimisations as you go along I spent a year without being able to play due to this not being done sooner. The game should not have been in that bad a shape for such a long period, and end game I'm sure if more effort is put in you will find a way getting some more horsepower out of the engine. Anyway, thanks, you quite literally could not have chosen a better time to fix the game....Got a lot of free time on my hands at the moment, living in Italy and all.

 

[Reman]

For the people who are having performance issues at the first of each new year, I'm fairly certain the problem is from lots of pops assimilating to psionic and/or synthetic. This is why it only shows up later in the game once ascension paths start getting taken. This slowly starts to fix itself as more species are assimilated, at least for when assimilation is occurring in human empires.

 

[KingAlamar]

ryzen 2600
RX 480 8gb
by 2350-2400 1000 star map 24 nations I'm looking at about 10 years every hr.
So to get to end game it would take 15++ hrs, and an entire game on fastest speed would take 25-30+ hrs.
Looking at several twitch videos, it seems this is fairly common, and around that time you also start getting huge hang ups, since all it seems they did was delay certain checks, instead of fixing the way it does it.
So while they may have increased performance, a bit, it is certainly not even close to enough to make the end game playable.

They seriously need to ditch the entire single pop representation and growth system they have because that is the problem and it wont go away until they do.

It's too bad that the game engine doesn't allow significantly beefier rigs to run better than older rigs. I could provide numbers but my computer is [technologically] 6 years older than yours is and we have roughly the same performance because my single-core performance is only trailing yours by about 10%.

While it will NEVER EVER happen I would just as soon go back to 1.9 and then re-add the DLCs and most goodies but keep MOST of the underlying systems from 1.9. Sure we'd need to kill that crazy whack-a-mole end-game issue and other obvious 1.9 problems but I would be happier and the go-forward path would be reasonably clear IMHO.

 

[ragehavoc]

It's too bad that the game engine doesn't allow significantly beefier rigs to run better than older rigs. I could provide numbers but my computer is [technologically] 6 years older than yours is and we have roughly the same performance because my single-core performance is only trailing yours by about 10%.

While it will NEVER EVER happen I would just as soon go back to 1.9 and then re-add the DLCs and most goodies but keep MOST of the underlying systems from 1.9. Sure we'd need to kill that crazy whack-a-mole end-game issue and other obvious 1.9 problems but I would be happier and the go-forward path would be reasonably clear IMHO.

Yes, the single core issue is still persistent, so it wont matter how beefy your computer is, it still wont run the game any different than someone on an 8-10 year old computer, which is the problem.

They dont actually need to go back to anything, they need to change how pops work, so instead of having singular pop representation, you have SPECIES with raw numbers of how many is in said species.
That way they can use the same checks, but it will check species instead of 100's of pops per planet. That alone would significantly reduce required computations.
It would also give a lot of advantages to the entire system, such as allowing you to make more than just one biological and one machine pop at a time, they could have a more realistic unemployment and criminal system to manage, as well as a real immigration system that can make you lose pops to other plants maybe even bordering empires.
Heck they could roll it into a DLC that focus's on empires cultures.
They really pigeonholed themselves with the single pop representation.

 

[El_Ploplo]

Yes, the single core issue is still persistent, so it wont matter how beefy your computer is, it still wont run the game any different than someone on an 8-10 year old computer, which is the problem.

I really don't understand why they can't add more parallelization to the code. The kind of computation needed for Stellaris are exactly the kind of computation that is quite easy to parallelize. Sure they are probably some difficulty with caching, I suspect that they use some kind of representation for the pop that it not cache obvious. They need to rework this and it will help incredibly.

I was planning to change my i5-3570k to a Ryzen 5 3600, do you know if it is useful for Stellaris and Hoi4 to run faster ?

 

[El_Ploplo]

It almost starts to freeze in Early Game. I end the game and it works fine again when I restart it...for some time. This is something new.
I don't know, either they need new programmers or they are in dire need of a new, better engine. How it's called, Clausewitz? I think something is wrong with that. It's not even Stellaris alone. There's no way I could play HoI4 til 1945, for example. LAAAG.

I don't think the engine is the real culprit here, the main issue with HOI4 is that all nations have more than 200 divisions near 1945, even the smallest one. Stellaris has exactly the same problem with it's pop system. At least, it makes the purifier species fun to play, trying to kill as much pop as possible to prevent lag

 

[KingAlamar]

Yes, the single core issue is still persistent, so it wont matter how beefy your computer is, it still wont run the game any different than someone on an 8-10 year old computer, which is the problem.

They dont actually need to go back to anything, they need to change how pops work, so instead of having singular pop representation, you have SPECIES with raw numbers of how many is in said species.
That way they can use the same checks, but it will check species instead of 100's of pops per planet. That alone would significantly reduce required computations.
It would also give a lot of advantages to the entire system, such as allowing you to make more than just one biological and one machine pop at a time, they could have a more realistic unemployment and criminal system to manage, as well as a real immigration system that can make you lose pops to other plants maybe even bordering empires.
Heck they could roll it into a DLC that focus's on empires cultures.
They really pigeonholed themselves with the single pop representation.

The reason I would like to "rewind time" [but add DLC] is that from 2.0 onwards a lot of things like Awakened Empires, Crisis, etc. was broken and haven't returned to what they once were. So it's not just for the performance / micro boost but I'd like to get back features of the game that simply aren't there any more. Of course that will NEVER EVER happen

 

[KingAlamar]

I don't think the engine is the real culprit here, the main issue with HOI4 is that all nations have more than 200 divisions near 1945, even the smallest one. Stellaris has exactly the same problem with it's pop system. At least, it makes the purifier species fun to play, trying to kill as much pop as possible to prevent lag

Maybe a better way to state it would be that the current engine is ill-suited for the current implementation of Stellaris? It would probably also be fair to say that the current engine does not SEEM to be as well equipped as it perhaps could be when it comes to systems with > 4 CPUs [threads].

Devil's Advocate: As someone that used to do multi-tasking [OSs] professionally getting this right is a HARD problem. Run-of-the-mill coders will result in systems that are relatively poorly threaded. Oddly I would contend that it would take REALLY ELITE coders to do a "good" job at wringing out a majority of the horsepower of an engine.

The problem is that if you significantly change the game you may need to get those "really elite" [expensive] people back in to make sure that the new implementation didn't break anything ... once again a difficult problem to manage.

 

[ragehavoc]

I don't think the engine is the real culprit here, the main issue with HOI4 is that all nations have more than 200 divisions near 1945, even the smallest one. Stellaris has exactly the same problem with it's pop system. At least, it makes the purifier species fun to play, trying to kill as much pop as possible to prevent lag

I am pretty sure you just described, "problems with the engine". And yes, that is the main issue with many PDX games, their engine is very old, it is not made for modern gaming, which is why they need to spend so much time finagling with the code just to make it run, "ok".

Devil's Advocate: As someone that used to do multi-tasking [OSs] professionally getting this right is a HARD problem. Run-of-the-mill coders will result in systems that are relatively poorly threaded. Oddly I would contend that it would take REALLY ELITE coders to do a "good" job at wringing out a majority of the horsepower of an engine.

The problem is that if you significantly change the game you may need to get those "really elite" [expensive] people back in to make sure that the new implementation didn't break anything ... once again a difficult problem to manage.

I mean, wouldn't it be better to get people in that can actually fix/improve the game than to have games that hardly work and wont make a lot of money because of that?
Maybe PDX should hire someone with a business degree too, LOL

 

[El_Ploplo]

I mean, wouldn't it be better to get people in that can actually fix/improve the game than to have games that hardly work and wont make a lot of money because of that?
Maybe PDX should hire someone with a business degree too, LOL

Well the issue is that is easier to sell new features than to sell fix and optimizations unfortunately. You can't sell a dlc based of optimization even if it will be probably more useful than new features. That is an issue with all paradox games and the reason why they experiment with subscription in eu4 (or CK2 I don't remember).

 

[KingAlamar]

I mean, wouldn't it be better to get people in that can actually fix/improve the game than to have games that hardly work and wont make a lot of money because of that?
Maybe PDX should hire someone with a business degree too, LOL

I think the time for that has passed. We're talking about a Level-of-Effort that could be justified for Stellaris II + EU V + ... [I.E. their next-gen-games] but can't be as easily justified when you don't have multiple new games rolling out to take advantage of that nice engine.

I suspect that there is still some low-hanging-fruit [comparatively] so there's still hope for more improvements.

 

[exi123]

Well the issue is that is easier to sell new features than to sell fix and optimizations unfortunately. You can't sell a dlc based of optimization even if it will be probably more useful than new features. That is an issue with all paradox games and the reason why they experiment with subscription in eu4 (or CK2 I don't remember).

In would say this is true for new games. In the case of stellaris, which is now live for four years, its not. The general vibe of the steallris community since 2.2 was very concerned about exact these features/problems: performance and ai. It WAS broken. It needed to be fixed. With the major fixing of 2.6.x stellaris came back to the state of "playable as expected". For me it wasnt the content from Federations which sold the DLC to me, it was the fixing.

I dont call them improvements, because the game was actaully playable with certain settings pre 2.2 pretty well for me. And the AI was able to build up planets and empires into the lategame staying "equivalent" to the player, without mods.

 

[grommile]

why they can't add more parallelization to the code

Adding parallelism to a system is frequently harder (and only rarely easier) than designing an inherently parallel system for the same task from scratch would be.

 

[KingAlamar]

@exi123 : Good point. I only bought Lithoids & Federations because of those fixes.

Honestly might consider NOT buying again until I see some QoL updates or fixes to other things I'd like to see addressed [AEs, Crisis, other event fleets, etc.]

 

[Annihilat0r]

Thing is, the mp lag in stellaris is weird as it comes from problems with sync. Prior to 2.6.x everything was fine, you start the game in mp and normal speed is same like in sp, fastest is same speed like in sp etc. After 2.6.X even fastest is like slow in sp, when u look at calendar you can see day go like this: 1 (3 second break) 23456 (in 0.5 second) 7 (after 5 second) 8910 (in 1 sec) 11 (after 4 seconds) etc. Theres 0 smoothness and you can see when hovering around clock/calendar in mp that other players are lagging 1-3 days behind host, thing is, in mp u can be max 3 days behind host, then host needs to wait for all to catchup which creates that artificial lag. So it not real lag performance wise.

We had the same problem yesterday when playing on "stellaris_test". I suggested to restart our games and switch to the main branch, which fixed the issue and the game ran as smoothly as in SP. I don't know if Paradox' servers were broken yesterday - since the new MP backend runs on their servers right? Or maybe just restarting and reloading the game fixed it already.