Performance Megathread

[MasterChief]

With the announcement of the new Federation expansion, I was both excited and a little let down at the same time. On one hand I have been hoping for a more in depth macro-diplomatic model and it appears the new ideas on federations might do just that... excited However, the trailer and accompanying information is not very enlightening regarding whether diplomacy in general will get a rework. For example, being able to ignore closed boarders (with a steep diplomatic penalty or reaction by the offended regime), sanctions, requesting boarder regimes to scale down military build up near your boarder, demanding turn over of systems, archeological sites or tech etc. So, I went from excited to just mildly interested Then, I stumbled upon a gaming website interpretation of the trailer that indicated that with warfare and the internal politics of Stellaris fully fleshed out, it was time to delve into federations... internal politics fleshed out, REALLY There is still so much that can be done to improve internal politics! Like tying internal politics into ethics, authority and civics … depending on your choices your game may be more like CK2 in space, EU4 in space or more in line with more traditional 4x game experiences. Troublesome governors, internal plots, etc. could create the rich mid-game playing experience so many forum posters have talked about. So now I went from excited to slightly frustrated

 

[~Robbie]

https://www.glassdoor.co.uk/Reviews/Paradox-Interactive-Reviews-E1008028.htm

Guess that shows us the reason, why performance is so horrible and nobody is adressing it. Dev teams are heavily underpaid and understaffed. Which quite explains the whole picture. We still get creative and cool content, because the developers are highly commited and love the game. But there are not enough ressources to actually flesh old things out. Just read through a few of these reviews. It is sad. There is even one person claiming talking about "the top-grossing, highest priority project in the studio, and there were only 13 developers working on it", which actually could be Stellaris, not sure.

Also wants me to say thanks to the Devs who still try to give us an awesome game. They just can not. Love you guys.

Kind of funny that corporate can afford to pay a Chinese developer to make a scam mobile game but not their own employees. Well, it would be funny if it weren't so sad.

 

[Ambereyed]

However, the trailer and accompanying information is not very enlightening regarding whether diplomacy in general will get a rework.

We know there will be new resource called "favours", which will used to gain more "diplomatic weight" during votes in Galactic Senate. So yes, probably diplomacy will be reworked a bit.

 

[ashbery76]

I'm sure internal will be worked on in the future.

As for scaled down military build up near your boarder, demanding turn over of systems, archeological sites or tech etc.This would exploit the A.I that doesnt understand basic strategic goals.

 

[Gromit]

The CEO's guide to ruining your company's reputation within a year.
by Ebba Ljungerud

 

[Cruel_]

https://www.glassdoor.co.uk/Reviews/Paradox-Interactive-Reviews-E1008028.htm

Guess that shows us the reason, why performance is so horrible and nobody is adressing it. Dev teams are heavily underpaid and understaffed. Which quite explains the whole picture. We still get creative and cool content, because the developers are highly commited and love the game. But there are not enough ressources to actually flesh old things out. Just read through a few of these reviews. It is sad. There is even one person claiming talking about "the top-grossing, highest priority project in the studio, and there were only 13 developers working on it", which actually could be Stellaris, not sure.

Also wants me to say thanks to the Devs who still try to give us an awesome game. They just can not. Love you guys.

Wow

If you are offered a job at PDS QA, do not take it. The people of PDS QA are talented, friendly and hardworking. However, they are poorly treated and very badly paid, even when just compared to wages at PDS. They are overworked, with most doing the workload of two people. Their reports and quality verdicts are regularly ignored (ever wondered why most products launch with major issues? this is why). On top of all that, they are regularly disrespected, lied to and blamed for problems. When any of these issues are pointed out to anyone higher up, QA are told that actually all the problems are caused by their “attitude" or “toxicity".

 

[_Perun_]

https://www.glassdoor.co.uk/Reviews/Paradox-Interactive-Reviews-E1008028.htm

Guess that shows us the reason, why performance is so horrible and nobody is adressing it. Dev teams are heavily underpaid and understaffed. Which quite explains the whole picture. We still get creative and cool content, because the developers are highly commited and love the game. But there are not enough ressources to actually flesh old things out. Just read through a few of these reviews. It is sad. There is even one person claiming talking about "the top-grossing, highest priority project in the studio, and there were only 13 developers working on it", which actually could be Stellaris, not sure.

Also wants me to say thanks to the Devs who still try to give us an awesome game. They just can not. Love you guys.

Careful, it seems to be a very sensitive topic for PDX (no wonder). There were threads about it in the last few days and got deleted. No wonder again, though, its easier and more profitable to sweep the problem under a carpet than to reform the company. Ethics doesnt work in business and every company is a company in the first place.

 

[ImbaXenoSnipar]

Was anything about performance mentioned during PDXCon? Not buying any more DLCs until the performance is fixed, so I really hope they addressed that.

 

[MasterChief]

I'm sure internal will be worked on in the future.

As for scaled down military build up near your boarder, demanding turn over of systems, archeological sites or tech etc.This would exploit the A.I that doesnt understand basic strategic goals.

Perhaps your are right... the AI rework would be significant

 

[Janx14]

Overhauls happen in the patches, DLCs just add new things.

I'm not garunteeing the patch that goes live with federations will include what you want, I'm just saying Federations' focus doesn't perclude the changes you want.

Think the only dlc that radically changes the game if you dont have it installed is utopia, but even then alot of its stuff has been slowly baked into the game.

 

[Apollo1784]

Was anything about performance mentioned during PDXCon? Not buying any more DLCs until the performance is fixed, so I really hope they addressed that.

That's gunna be a no from paradox champ. We all get to buy new dlc for a broken game yay!

 

[Serenity84]

If they don't add internal politics and also more complex external interactions into the free patch, I'm done with the game. It's been missing from the start and I'm done waiting for it. Free patches are the place for stuff like that however, and DLC is always accompanied by substantial free changes. So there is certainly a good changce that they'll do some things there. They can easily have the Space UN as a paid feature and make other more basic interactions free. But nothing else has been announced yet.
I'm somewhat hopeful, but not too much given the tortured development and changing visions of the game so far.

but even then alot of its stuff has been slowly baked into the game.

Yes. They realized that locking ascension perks into the DLC wasn't a good idea. So they made the basic ones free and only have specialized, thematic ones in the DLC now

 

[The Founder]

Multi core is an overrated marketing ploy created because it is no longer possible to make faster processors due to heat issues with Silicon which is why research is done to switch over to carbon which have much better heat properties.

The reason why multi core don't work anywhere as well as you may think is because of something called Amdahl's law which basically say if something can't be run completely parallell (and I suspect games like Stellaris have alot of those stuff) the performance gain from additional cores will drop sharply.

For games it is a loose/loose. The core of all game programming, is the game loop:
https://en.wikipedia.org/wiki/Game_programming#Game_structure
And aside from the ovious stuff (like file loading), there is nothing that will benefit from Multitasking. Not without a massive rework or various design tradeoffs for the underlying mechanic.

That is why I mentioned the "multithread calculating the Fibbonacci sequence". That is what the game loop boils down too.

The reason why Stellaris have become slow is because they have added more stuff that Drains performance such as complicated pathfinding for trade routes, if they used a simpler solution or better optimization, performance would start to return.

The main drain on pathfinding is ironically gateways. And it is harder then reallife pathfinding, because those guys do not have to deal with 1:N teleporters all over the map.
I did start a discussion about this ages ago including a pretty simple way to fix this - if we could live without player defined 1:N gateway networks:
https://forum.paradoxplaza.com/forum/index.php?threads/pathfinding-and-bypasses.1180725/

My argument is that Paradox should develop their games with the limitations of their engine in mind.

So only give us the 250-500 Star range? Sorry, but the Marketing Department really disagreed with that.

I don't know:

• The source code for Stellaris.
• How to assess how parrellized it is (optmized for multi-core processing).
• How to make it so, or even how to assess at what percentage it is.

We do not need to know the Source Code. This is really just 101 of Multitasking: It is not a magic bullet.
Use it for the wrong problem, and you actually end up making the calculation slower because you just added a ton of overhead.

 

[krios41]

As for scaled down military build up near your boarder, demanding turn over of systems, archeological sites or tech etc.This would exploit the A.I that doesnt understand basic strategic goals.

could be implemented as a MP featuref- for use against humans only? idek

 

[Acacias]

In the stream there were hints that a new engine was in the works, as Daniel repeatedly referred to stellaris being "in an old engine" when askedabout hotkeys.
Plus they are making CK3, I doubt that will be in the old engine

 

[Ecervele]

In the stream there were hints that a new engine was in the works, as Daniel repeatedly referred to stellaris being "in an old engine" when askedabout hotkeys.
Plus they are making CK3, I doubt that will be in the old engine

Don't underestimate Paradox...

 

[Ecervele]

Your computer with the federation DLC

 

[lovenotwar]

Wow

If you are offered a job at PDS QA, do not take it. The people of PDS QA are talented, friendly and hardworking. However, they are poorly treated and very badly paid, even when just compared to wages at PDS. They are overworked, with most doing the workload of two people. Their reports and quality verdicts are regularly ignored (ever wondered why most products launch with major issues? this is why). On top of all that, they are regularly disrespected, lied to and blamed for problems. When any of these issues are pointed out to anyone higher up, QA are told that actually all the problems are caused by their “attitude" or “toxicity".

This seems to be their go to for anything negative said about the company, every one of those comments blame management for the environment, which doesn't surprise me, given some of the things I have heard them say to even their own customers.
Sounds like the company is on the verge of collapsing at any time if it wasn't for their underpaid workers staying with them for god knows why, a shame really, they should all go on strike.

 

[alexti]

Because despite what CPU Manufacturers hyped, Multithreading is not a magical "solve all performance problem" bullet. It works for some very specific problems. And the kind of problems of wich games usually have very few.

It works well for most practical problems. For example, in Stellaris, every AI empire can evaluate their decisions in parallel, and every planet can perform pop/job matching independently. It's just a matter of programmers writing suitable code.

If you know the magical way to Multithread the calculation of the Fibbonacci sequence (and nope, storing the sequence does not count as calculation), tell us.

I am not sure about magical, but here is a practical way. Let's say you need to compute Fibonacci sequence from F(n) to F(m). If you have p cores, split the interval (n;m) into p equal segments. Let's say each part starts at s_i number. For each part compute F(s_i) and F(s_i+1) via characteristic polynomial (for Fibonacci sequence it would be something like F(s)=c*(a^s-b^s), where a,b,c are constants). Then compute all other elements of the segment via linear recurrence formula. Each segment of the sequence can be computed independently so you can use all p cores that are available. With the current processors you could also use SIMD instructions to compute several segments on the same core simultaneously.

A lot of computations are like that - they can be done in parallel, but it won't happen magically, you need to think a bit before writing the code.

Anything that was a bottleneck and could be multithreaded has been - usually with a massive redesign to the mechanic. Just look at Migration, wich was redesigned twice for MT support

There is no evidence that pop-job matching is parallelized and several players have run experiments (for example earlier in the thread) showing that pop-job matching is a large bottleneck. The main non-parallelizable bottleneck is rendering, which in theory is parallelizable, but it's not under PDX control until they switch to Vulcan.

 

[alexti]

Multi core is an overrated marketing ploy created because it is no longer possible to make faster processors due to heat issues with Silicon which is why research is done to switch over to carbon which have much better heat properties.

The reason why multi core don't work anywhere as well as you may think is because of something called Amdahl's law which basically say if something can't be run completely parallell (and I suspect games like Stellaris have alot of those stuff) the performance gain from additional cores will drop sharply.

Overall the gains from new CPU have been limited, most of the major development have been in GPU which is specialized in doing stuff parallell such as graphics while CPU technology have largely stagnated because of reaching physical limits, now it is mostly about quanity, adding more transistors instead of better ones.

It seems you are falling to a marketing ploy of GPU makers What GPU is doing is throwing a massive number of simple cores at the problem. To a larger degree than general purpose CPU do. There isn't that much difference between GPU and CPU. GPUs are just more narrowly specialized and more parallel. The reason you don't see much progress with CPUs is that generally Intel wasn't bringing the more powerful chips to the consumer market. It might somewhat change now though - it seems they are about to release 18-core i9-10980. Of course, this is still mostly about widening the computational bandwidth. Not that it's going to help Stellaris though (without some code changes).

I don't know:

• The source code for Stellaris.
• How to assess how parrellized it is (optmized for multi-core processing).
• How to make it so, or even how to assess at what percentage it is.

I don't think anyone else here does either, it could be anything, chances are that if there's room for even a 25% increase it'd be dramatic though. This is a problem with the average issue of multi-core processors, that is, how optimized tasks are on average out there. And chances are, it ain't much.

You can easily measure how much parallelized it is and what percentage of cores it uses by using performance monitoring tool of your choice. For example, see some results earlier in the thread.

My experience from talking to most programmers is they don't even want to touch parallel programming because it's the sane approach.

That's funny and there's some truth to it - some programmers are always willing to go an extra mile to avoid using the sane approach