System requirements for no lag in 1K stars galaxies?

[frenchyvinnie]

Hello everyone, this is my first time posting on this forum, so if my question is misplaced or has already been answered, then my apologies.

So my question is: What system requirements would allow you to play on a 1000 star galaxy with no lag/stutter, even in late games?

My current system is as follows:

• Strix GTX970
• I7 6700K @3.2GHz (haven't overclocked it yet)
• 16GB DDR3 of RAM

I was playing on a 1000 star map and didn't run into any problems for the first ~150 years. Then the War in Heaven happened and everything went to shit lol. Stuttering galore and my PC's fans are rotating louder than an A380 on takeoff.

Is there a way to play on galaxies of this size with no stutter, even in late games? If so, what should I upgrade first with my current setup?

 

[WolliMW]

Thats weird.
My System:
i5 4690K
GTX 1070
8 GB Ram

I got lags at around 200-300 years in game. But i never got so much lag that i cant play anymore.
I wont observe any fleets at that point so im fine
I think there is a mod somewhere out there tuning down fleet size so there will be fewer lag

 

[The Founder]

I7 6700K @3.2GHz (haven't overclocked it yet)

It is a I7. It propably overclocks automatically.

Games usually need single-thread performance. The current generation of Multicore CPU are not ideal for them, and Stellaris is no exception.
However having cooling designed for 4 or 8 cores at full load, means one could opearte one core far beyond full load if the rest is twidling virtual thumbs anyway. That is why any modern multicore CPU will ship with a build-in, fully automatic overclocking.

 

[mackau]

Mid-late & Late-game pausing is inevitable on all but the smallest galaxy sizes because the game works almost exclusively on a single core and can run the amount of fleets/pops/objects/planets etc under any current CPU.

 

[Drowe]

I wonder how much work it would be to adapt the Clausewitz engine for multi core usage. And how much performance would improve if it was. It isn't like parallel computing is easy or quite as good for performance as one might think.

Parallel computing introduces points of failure that simply can't happen in nonparallel programs, which in turn requires much more effort to avoid bugs. And even then some bugs may not be reliably reproducible and only occur randomly, making it much more difficult to track them down.

Also some things can't be processed any faster in parallel than in sequence and the additional care, that needs to be taken to avoid bugs, could actually slow the program down instead of speeding it up.

 

[frenchyvinnie]

Mid-late & Late-game pausing is inevitable on all but the smallest galaxy sizes because the game works almost exclusively on a single core and can run the amount of fleets/pops/objects/planets etc under any current CPU.

Well crap. I hoped there was a way to keep my game smooth in late games by upgrading something. But if the flaw lies with the game's engine, then it's hopeless.

 

[The Founder]

I wonder how much work it would be to adapt the Clausewitz engine for multi core usage. And how much performance would improve if it was. It isn't like parallel computing is easy or quite as good for performance as one might think.

Parallel computing introduces points of failure that simply can't happen in nonparallel programs, which in turn requires much more effort to avoid bugs. And even then some bugs may not be reliably reproducible and only occur randomly, making it much more difficult to track them down.

Also some things can't be processed any faster in parallel than in sequence and the additional care, that needs to be taken to avoid bugs, could actually slow the program down instead of speeding it up.

First they would need a problem that can even be multithreaded. Multithreading has to choose it's problems carefully.

That being said, at least the AI is running multithreaded. Something else was moved to multithreading too with 1.5.
But in the end the main game processing (the game tick) must run in a single thread. There is just no way around it. And it has to wait if any other thread takes to long to answer.

 

[FrosT37]

System requirements for no lag in 1K stars galaxies?

Answer: these system requirements do not exist yet

 

[Teutonizer]

That's the thing with almost every PDX game (it's about current, older ones can run perfectly on new hardware, EU3 for example), single-core usage and all that

System requirements for no lag in 1K stars galaxies?

Answer: these system requirements do not exist yet

Well, you can have no lag even in 5k...for first 10 minutes

 

[Bryartuck]

I wonder how much work it would be to adapt the Clausewitz engine for multi core usage. And how much performance would improve if it was. It isn't like parallel computing is easy or quite as good for performance as one might think.

They already do thread as much as they can to improve performance. However, most of the processing occurs inside the main loop, which is single threaded by its nature.

 

[mathers]

First they would need a problem that can even be multithreaded. Multithreading has to choose it's problems carefully.

That being said, at least the AI is running multithreaded. Something else was moved to multithreading too with 1.5.
But in the end the main game processing (the game tick) must run in a single thread. There is just no way around it. And it has to wait if any other thread takes to long to answer.

Not necessarily:4 Core runs their own ticks for 1/4 of galaxy each, and they sync up less frequently.

 

[The Founder]

Not necessarily:4 Core runs their own ticks for 1/4 of galaxy each, and they sync up less frequently.

That runs into issues on the borders. They actually planend something like that for Planetary Annihilation, but it failed.
The moment two threads have to interact, you got another bottleneck. And the bottleneck will be just looking if they have to interact.

I am not entirely certain how minecraft solved it. But there chunks are pretty smal 64 blocks cubed, I think?

 

[mathers]

That runs into issues on the borders. They actually planend something like that for Planetary Annihilation, but it failed.
The moment two threads have to interact, you got another bottleneck. And the bottleneck will be just looking if they have to interact.

I am not entirely certain how minecraft solved it. But there chunks are pretty smal 64 blocks cubed, I think?

All need to know the galaxy and use only pseudorandom in ai therefore entities at border calculated by all impacted segment. Close to border is precalculatable

 

[terrycloth]

All need to know the galaxy and use only pseudorandom in ai therefore entities at border calculated by all impacted segment. Close to border is precalculatable

Keeping that in sync sounds like a nightmare bug farm.

 

[Drowe]

First they would need a problem that can even be multithreaded. Multithreading has to choose it's problems carefully.

That being said, at least the AI is running multithreaded. Something else was moved to multithreading too with 1.5.
But in the end the main game processing (the game tick) must run in a single thread. There is just no way around it. And it has to wait if any other thread takes to long to answer.

Multi-threading is not quite the same as parallel computing, many programs run multiple threads and still only use a single core. A simple P2P chat application needs at least 2 threads, but they don't need to run on separate cores and wouldn't gain anything from doing so.

That being said, my question was more a hypothetical one. Could the engine be adapted to actually require multiple cores, and if so would that actually make the game faster. That goes beyond just doing some things in parallel and requires a different approach altogether. However I can't say if such an approach would even be possible for Stellaris, at least not without some more understanding of the code. And I did mention that not all problems can be solved faster by parallel computing.

 

[The Founder]

Could the engine be adapted to actually require multiple cores, and if so would that actually make the game faster.

The launcher alone has 16 Threads running. The game itself clocks in 30-40.
But most of them are propably just leftover diskreader threads or ones waiting for specific action to do anything at all. And even that will be far from a CPU bound operation.

You can start as many threads as you want. If they do not require a fair share of the CPU time (that the main thread no longer has to deal with), it will not speed up the whole process.

 

[mathers]

Keeping that in sync sounds like a nightmare bug farm.

Agree.

I think Stellaris is fine as it is. I prefer them work on extensions instead. However my engineer side says it would be possible to refactor it in a way that no need of central core.

 

[Alblaka]

I wonder how much work it would be to adapt the Clausewitz engine for multi core usage. And how much performance would improve if it was. It isn't like parallel computing is easy or quite as good for performance as one might think.

Parallel computing introduces points of failure that simply can't happen in nonparallel programs, which in turn requires much more effort to avoid bugs. And even then some bugs may not be reliably reproducible and only occur randomly, making it much more difficult to track them down.

Also some things can't be processed any faster in parallel than in sequence and the additional care, that needs to be taken to avoid bugs, could actually slow the program down instead of speeding it up.

FYI, devs already confirmed (3+ times in the one or other 'why doesnt Stellaris have multithreading yet') threads that Stellaris DOES use multi-threading. Just that there are a few critical core processes that cannot be multi-threaded, and those are the ones taking up the most time anyways.

As to how much of that is actual parallel computing, I can't tell, might be worth poking that question at them.

 

[ashbery76]

Truth be told medium should be maximum size map.The old engine cannot handle more.

 

[Summin Cool]

I didn't see lag up to 2400ish with a 1000 star 73 empire mess. I recall stopping playing at 2420ish

I did stress test this and my rig starts to lag around 700 fleet power in system (in a seperate game).