Home Trending Latest threads New posts Developer diaries

Performance Megathread

Jump to latest Follow Reply
  • We have updated our Community Code of Conduct. Please read through the new rules for the forum that are an integral part of Paradox Interactive’s User Agreement.
Stellaris 52768 Bug Reports 33939 Console edition 1286 Suggestions 20844 Tech Support 3128 Multiplayer 390 User Mods 4759 Stellaris AAR (After Action Reports)
Show only dev responses
alexti said:
When you think about it this doesn't seem to an inherent problem in Stellaris. For example, pop evaluation on each planet could be done in parallel. In most (if not all) scenarios there's no contention for any data at all.
Click to expand...

I agree. The parallelisation of many functions could be improved. However, refer to an earlier post where I talked about engine updating and business realities. There's only so far @Moah and co. can get under the bonnet, before they're likely treading on the toes of the team doubtless working on ClauseWitz2whatever.

I doubt he can comment, but I pinged Moah to see if he wanted to clarify regarding this.
 
so, 2.4.x came out, and not noticing any real improvements in performance, 2500 IG and the game ticks approx 15-25 times a min with 0.25x planets and gates and 15 empires, and getting a nice steady 20 FPS, because people DO NOT KNOW HOW TO SPLIT THE RENDERING THREAD FROM THE SIMULATION THREAD (ffs, just think how much of a perf improvement would result if the main thread did not have to deal with rendering(may not be true, but FPS being heavily influenced by single thread load is a red flag))
anyone else noticing a difference?
 
Game year 2350, 2.4.1, 18 empires, 1.25x wormholes and gates, 1x habitable planets, medium galaxy (600 stars). Khan just woke up. Game ticks at 75 days per 30 seconds passed. For me, this is a marked improvement over the previous versions where on max speed (such as this test) game was already chugging much slower, at more like 30-40 days per 30 seconds. I honestly couldn't care less about FPS, but if you are concerned with the rendering impacting performance, just turn all the graphics all the way down (and see if it helps, and report back).

Edit: it's now 2384, game still ticks along at almost 60 days per 30 seconds.
Edit2: for more data points and anyone wondering, 2394, 47 days per 30 seconds.
Edit3: even more data points: 2400, 47 days per 30 seconds. (This is still faster than second-fastest ingame speed, clocking in at 33 days per 30 seconds)
 
Last edited:
oh, well mabey im asking too much expecting a overclocked GTX 1060 6GB with 12 CPU threads to push 60FPS in this game >_> (and a decent tick rate thank you very much) (or have an actual playable game in the endgame rather than this slideshow shennanigans)
 
Last edited:
Quick question: how many empires in your game do you reckon have access to the xenocompatibility AP (as in, either have it themselves, or have a migration treaty with someone who does)? From what I've observed myself, the game performance nosedives once the species count goes way up, even if the pop count doesn't really.

"The Fanatic Purifiers were right all along!"
 
Olterin said:
I honestly couldn't care less about FPS, but if you are concerned with the rendering impacting performance, just turn all the graphics all the way down (and see if it helps, and report back).
Click to expand...

AFAIK it never did anything to people having good enough GPUs. Will only help if GPU is struggling while being 100% loaded. The problem is that rendering and simulation being performed in the same thread, one of them overloading the core executing this thread drags the other down. Even if lowering graphics helped, that's a problem that needs solving.

About xenocomp, I did a few tests about species count and observed pretty much the same. Bio-engineering also is a problem as AIs will spam subspecies. By removing both and also removing habitats from the game, and using the EDAI mod cutting on pop calculations, I manage to get around 10s for 30 days in 2400 on a 1000 stars with 15 starting empires (default setting). That's with a i7 6850k 6 cores OC@4.2GHz. Empire count is also a big factor. The same game with 30 empires (max setting) is getting 15s for the same 30 days, same date.
 
oh hell yeah, just switching between tabs in the species menu causes a good 20 second lag spike while it loads (and not to mention, but several species have about 7-20(yes i have seen that many) variations, for about 18 different species
so, lets say on average there are 15 variations of each species for 20 species, that is 300 different variations just sitting around, then add the fact that the pop system has to run through and individually manage each one of those (i believe the current system groups based on variation+job) so that up to 300 different groups for each fucking pop job there is, on each planet (up to, realistic numbers usually dont go past 2-5 groups per job) and how many planets are there in a usual play-through? with how many jobs? (just a thought)
 
Last edited:
FR4CT4LH3X said:
oh hell yeah, just switching between tabs in the species menu causes a good 20 second lag spike while it loads (and not to mention, but several species have about 7-20(yes i have seen that many) variations, for about 18 different species
so, lets say on average there are 15 variations of each species for 20 species, that is 300 different variations just sitting around, then add the fact that the pop system has to run through and individually manage each one of those (i believe the current system groups based on variation+job) so that up to 300 different groups for each fucking pop job there is, on each planet (up to, realistic numbers usually dont go past 2-5 groups per job) and how many planets are there in a usual play-through? with how many jobs? (just a thought)
Click to expand...
That's why I'm an advocate for a new system where number of pops would be drastically cut unless whole approach with pop as single entity would be abandoned.
 
Anyone tracking performance improvements of the most recent patch?

At this point I feel like Paradox just adds on "blah blah improved performance of something" as a ruse to get me to come back and try a new game before giving up when performance is still awful end game. I'd love to have confirmation of a meaningful improvement before doing so.
 
CrazyZombie said:
That's why I'm an advocate for a new system where number of pops would be drastically cut unless whole approach with pop as single entity would be abandoned.
Click to expand...

From what I tested and from what I read from the descriptions of the mods I use, the big offender seems to really be number of sub-species, total number of pops or total number of planets only playing a marginal role in this in the end. The big thing is having planets with lots of different species on them thus needing lots of job evaluations, where single species' planets, even crowded, will be a breeze. Xenocomp, bio-ascension, primitives, migration pacts, xenos integration through war, numerous starting empires, all indirect lag source through pop/job eval.

FR4CT4LH3X said:
or the AI does not spam 15 different bloddy variations?
Click to expand...

That comes from bio-ascension AP. Apparently the AI once it bio-ascended just keeps churning out new useless sub-species for the sake of genemodding, also killing its society research in the process. Regular gene tailoring seems to stop things when they run out of trait points, but the AP, granting ability to remove good traits, fires an infinite loop of genemodding. Here the description of the mod I use for preventing it : https://steamcommunity.com/sharedfiles/filedetails/?id=1801814489
 
wait, so it ONLY occurs once it hits the perk? and not any of its prerequisites, it just keeps making new variants? what in the absolute fuck is paradox thinking on not putting some sort of control on that, what the fuck! (and just a complete general oversight on it only being with that one perk?)
 
I keep telling people, the best thing to do would be for Paradox to get rid of pops and jobs entirely; this obsession with keeping something that doesn't work properly just because that's how it's always been done is only hindering the product.
 
Blayze said:
I keep telling people, the best thing to do would be for Paradox to get rid of pops and jobs entirely; this obsession with keeping something that doesn't work properly just because that's how it's always been done is only hindering the product.
Click to expand...
Well, it is definitely a solid option. Problem is - how to explain to PDX that "simplified but optimized" game is better than "micro-intensive hell that is also unplayable due to lag".
 
CrazyZombie said:
Well, it is definitely a solid option. Problem is - how to explain to PDX that "simplified but optimized" game is better than "micro-intensive hell that is also unplayable due to lag".
Click to expand...

Easy: I haven't bought anything since they duped me with Megacorp, so if they don't fix the product they sold me, there's no point me giving them any future money for products that won't even work for me.
 
FR4CT4LH3X said:
wait, so it ONLY occurs once it hits the perk? and not any of its prerequisites, it just keeps making new variants? what in the absolute fuck is paradox thinking on not putting some sort of control on that, what the fuck! (and just a complete general oversight on it only being with that one perk?)
Click to expand...

It can't on the prerequisites, good traits are not removable. Once all pops are modified there's nothing left to do. It's auto-limiting. The final bio-ascension perk was implemented after and removes that limitation, apparently without paying attention to the fact it was a keystone of the AI not going nuts with this. Then I guess nobody never ever took a look at it again. So the oversight is that they didn't think to do anything at all and it was all good except for this one perk.
 
Dëzaël said:
It can't on the prerequisites, good traits are not removable. Once all pops are modified there's nothing left to do. It's auto-limiting. The final bio-ascension perk was implemented after and removes that limitation, apparently without paying attention to the fact it was a keystone of the AI not going nuts with this. Then I guess nobody never ever took a look at it again. So the oversight is that they didn't think to do anything at all and it was all good except for this one perk.
Click to expand...
It's kind of interesting because robots don't have this limitation to begin with. Once you unlock robomodding (Machine or Mechanical, doesn't matter), trait points and the amount of traits are the only limits imposed. The AI seems to only care about the bio pops in this regard; I've noticed they seem to leave robots alone.
 
On 2.4, I played a game into 2400s on default settings. It's definitely better but still a long way to go. Gone from "terrible" to "bad" basically.

I'd say that minimum expectations should be that on default settings on typical hardware that a complete game can be played without performance being a noticeable drag on the enjoyment. Assuming they don't make things worse again, it feels to me like it's going to take 1-2 years for the game to reach those minimum standards.

And it's not like there's no other major issues - mid/late game is still a drag.
 
You must log in or register to reply here.
Top Bottom
Back