Performance Megathread

[Ertest]

The Ryzen mobiles could shed some light on the effects of chiplet vs monolithic die latency since they use a monolithic die. Anandtech compares the latency between a Ryzen 3950X vs a 4900H and its substantal. Comparing a Ryzen 3100 vs a 3300 would also be interesting, one uses two dice (3100) vs one die (3300x).

Relevant links below.

AMD’s Mobile Revival: Redefining the Notebook Business with the Ryzen 9 4900HS (A Review)

www.anandtech.com

The AMD Ryzen 3 3300X and 3100 CPU Review: A Budget Gaming Bonanza

www.anandtech.com

until stellaris is fully parallelized to take advantage of all the processor cores, it is useless to solve latency

so far it's like driving a car with an eight-cylinder engine on one cylinder and thinking that changing the color of the car will increase acceleration

 

[James Fire]

The Ryzen mobiles could shed some light on the effects of chiplet vs monolithic die latency since they use a monolithic die. Anandtech compares the latency between a Ryzen 3950X vs a 4900H and its substantal. Comparing a Ryzen 3100 vs a 3300 would also be interesting, one uses two dice (3100) vs one die (3300x).

Relevant links below.

AMD’s Mobile Revival: Redefining the Notebook Business with the Ryzen 9 4900HS (A Review)

www.anandtech.com

The AMD Ryzen 3 3300X and 3100 CPU Review: A Budget Gaming Bonanza

www.anandtech.com

There are no monolithic ryzen cpus. Monolithic is a specific term that refers to having everything on a single die. No ryzen chip has this, since, at the very least, the IO is a separate die from the cores/cache.

 

[exi123]

While cpus are getting faster and increase the power per clock, i guess a game like Stellaris and other mostly single threaded will only profit if you upgrade the structure under the current cpu's.

The next step forward will be the introduction of ddr5, paired with new chipsets and new series of cpus next year. Until then, nothing on the market will give you a magic fluent gaming experience in this game. And even on these extreme fast machines the engine will lag in the lategame, because the engine cant handle the game... It isnt just slow, it always has these lag spikes, and these will never really be gone when not fixed by the devs.

 

[Ertest]

if Stellaris only works on one core, it will always be a performance issue

see for example Factorio - a good example of how to use multithreading - https://factorio.com/

 

[exi123]

if Stellaris only works on one core, it will always be a performance issue

see for example Factorio - a good example of how to use multithreading - https://factorio.com/

Factorio is an absolute bad example for multithreading. The devs have the same problem as the Stellaris ones: there needs to be to much calculated in the mainthread in realtime. They improved their performance drastically over the years, but not (only) with multithreading. Read their FFFs and you will see that they optimize mostly around their entities. A good example is how they redone the calculation of items moving on belts with a new algorhythm. This alone saved up to 50% of your processing power in belt heavy bases (and i bet optimizations like this are also possible in Stellaris, but it needs time and passion to do).

I always reach the maximal processing power of my pc with that game, when the ups start to shrink the game becomes unfun which is even worse because it is open end and unlimited. A play in Stellaris comes to an end at some point...

 

[Ertest]

Factorial is a good example of how to make modern games for modern processors, yes devs solved the same problems as in Stellaris but the result is completely different, there is no single core running at 100% load as in Stellaris

 

[CPR]

This has come up before but I think there's a bit too much focus on multithreading in this topic.

The raw scale of the problem to be solved (eg number of pops) is not that large. It doesn't seem to me that intrinsically complicated to calculate the basic economy from day to day. I see no obvious reason why they couldn't get 10x more single threaded performance, for example. Ultimately though, it doesn't matter how - we just need better performance, particularly better scalability (late games and bigger galaxies).

My general impression is that it's more like the management don't see performance as being a real issue. At least, not one worth dedicating significant resources to. It wouldn't surprise me to learn if the actual devs are frustrated that they can't spend the time they want on it. This is why I've taken the angle of more trying to put pressure on the management team, eg with surveys etc but I've not seen much support for that.

A shame, since Stellaris could be such a good game if they just focused on polishing what they have. I don't have an intrinsic problem with the new features they've added - they generally feel like the right direction but it's the implementation that's the let down.

For reference, I've been playing Civ 6 recently since I couldn't even find the motivation to try Stellaris 2.7. Never played it before or any previous version. I'm not that impressed. Performance is poor even at the start (too slow with default settings) though fortunately it doesn't seem to get particularly worse later on. Reliability is very poor (I'm lucky if I can go a day without it crashing on me or randomly becoming unusable). Game play and strategy side feels rather flat as it seems to be trying too hard to be "balanced". There is a steep learning curve but it doesn't especially feel like you get the appropriate rewards for investing the time to learn how to maximise your results.

 

[Ertest]

My general impression is that it's more like the management don't see performance as being a real issue.

I agree, If enough people buy it, they don't see a problem

 

[Anemetius]

Performance will never be fixed. Paradox has no technology or know-how. They are still stuck in early 2000.

 

[Xmostera]

I wonder if Developer checking their pinned posts everyday or once in a week, do they notice that we desperately need an update/rescale about performance? but again, where is everybody? *Fermi paradox*

 

[KingAlamar]

As I've mentioned a dozen times, feels like a million, if the bean counters do not believe that addressing performance will net them MORE money in the relatively short term then they'll always put off performance in favor of something that WILL earn them more short-term money.

Note: I do suspect further optimizations are possible. I don't think we will see anything more parallel unless the underlying system of how Stellaris' variables / buffs / debuffs / effects influence each other is changed in order to make parallel computing something that "falls out naturally" as opposed to a long series of SERIAL computations.

 

[CPR]

Performance will never be fixed. Paradox has no technology or know-how. They are still stuck in early 2000.

Without being on the inside, it's hard to know why certain decisions were made.

Considering it's a completely new game, I've no idea why the team decided to go with an old game engine with little/poor considerations for modern CPUs. Did the devs consider it to be a non issue or were they concerned but overridden by management for cost/time reasons?

It also feels like the devs / QA team weren't able to put in place from the start an extensive automated testing system, which probably explains why a lot of bugs resurface. Is that because such things are boring or because management set extremely tight deadlines that didn't make such things practical?

I could go on but based on my real world experiences as a developer, it's hard to know where to point the finger. There's certainly plenty of poor developers out there. But it's also very rare that developers are given the opportunity to follow "best practices" because management want everything done yesterday and are rarely concerned about the long term enough to do things properly in the short term. Doing things properly takes a long of time up-front. But for multi-year projects that can save you a lot of time in the long term.

From various comments, I certainly get the impression that the devs/QA are under-resourced for the scale of the project. It's hard to evaluate how competent they truly are outside of that because most of the "dev diaries" are about content rather than programming itself. There's been some times when I've been rather unimpressed with some of their comments but without knowing the problem in detail it's hard to truly judge.

(Just to be clear, sometimes the management team are right in that getting things done quickly/cheaply is the only way to have a viable product but that doesn't excuse never fixing those underlying issues once you start making a profit)

 

[Ertest]

if management receives such feedback then they do not know that there is a performance problem

Stellaris Survey Insights 2020

Hello Stellaris Fans! We're finishing up some things before we go on vacation, and from the results of our latest CRM survey, we have a few insights we thought our Community might find interesting! This survey was taken during March and April...

forum.paradoxplaza.com

 

[KingAlamar]

I think Stellaris' main problem arises from the DEVs naturally wanting to present systems / interactions that seem cool without taking into account what the performance / gameplay impacts of those decisions are. When that happens you often wind up with large amounts of computations that need to be done in a SERIAL fashion and are not well disposed towards taking advantage of modern hardware.

What needs to change is that Stellaris [Stellaris 2?] needs to be redesigned from the top down with systems in place that NATURALLY are easy to run in parallel with relatively low overhead. Then at least one lead-architect needs to be in a position to flat-out-veto changes that would introduce lots of 'serial computational overhead' or similar -- more specifically the lead-architect needs to work with the 'creative DEVs' in order to tweak their ideas so they are not DESTRUCTIVE to the overall performance of the game.

I suspect that element of authority & oversight is what's been missing for a long time ... not to mention it doesn't help when the vision of what Stellaris is supposed to be "radically" shifts every few years or so.

 

[Varileztra]

As a matter of interest, why in Stellaris which is not an FPS/TPS or RT action game, do you need more than 60fps? I am asking so I can understand your point of view

Bit late with a response, but i can tell you that: It smoothes all the movements further, making it pleasant to the eye.
Put your monitor to 120/144 hz, whatever it supports, move your mouse and then set it to 60 anddo the same. The difference is insane and once you're used to having 100+, going below becomes painful

 

[KingAlamar]

Bit late with a response, but i can tell you that: It smoothes all the movements further, making it pleasant to the eye.
Put your monitor to 120/144 hz, whatever it supports, move your mouse and then set it to 60 anddo the same. The difference is insane and once you're used to having 100+, going below becomes painful

In addition it's not just the issue of being used to 120+ Hz. Another issue is that the inter-frame timing isn't particularly consistent either which adds to what some people are referring to as herky-jerky motion. Depending on how sensitive you are to motion issues you can pick it up on 60Hz monitors easily.

Personally I've altered the way I interact with the game, where I keep my eyes focused, etc. to minimize the impact. To me it's not a huge deal but I could see it being a huge deal for folks that don't make the same sacrifices I do.

 

[UltimateTobi]

going below becomes painful

I can't go below 144hz again... I notice the difference even though technically you don't see more than 60 FPS. (or even 30?).

 

[GnoSIS]

So you heard it guys, performance will be improved in "incremental steps", which means this thread will reach at least double the page count, as this drags on for years.

They did hav a dev work on the loading speed problem all summer, to provide a 100% speedup, but no such thing was or will be done for the big issue.

And guess what, in the current implementation each pop holds a global lock to update empire stockpiles/production - yes it's THAT BAD!

 

[trybald]

So you heard it guys, performance will be improved in "incremental steps", which means this thread will reach at least double the page count, as this drags on for years.

They did hav a dev work on the loading speed problem all summer, to provide a 100% speedup, but no such thing was or will be done for the big issue.

And guess what, in the current implementation each pop holds a global lock to update empire stockpiles/production - yes it's THAT BAD!

Why bother with performance if you can just churn one DLC after another? The game is already like 4 years old. There's no economic gain in allocating resources to deal with persistent issues. It won't bring any new customers and old suckers customers who already bought it have gone into the sunken cost fallacy so they can be easily herded back with promises of minor improvements.

 

[Apollo1784]

It's literally incredible that they were eager to announce a decrease in loading screen times. I've never seen a company maybe outside of EA that has ever been so fundamentally incompetent and disconnected from their consumer base. And everytime they announce that nothing about performance will be fixed at all the massive horde of die hard brand-loyalists come out in full swing defending everything paradox does. This company could sell them a plate of dog shit and they would ask for more. Not to mention the almost complete ignoring of this thread by the Devs AND the low-key unannounced censorship that goes on in these forums make for one hell of a negative customer experience. And im sure this post will get censored too.