Yup, Dev Diary #120:
Also the 2.2 patchnotes mention:
"Numerous other bugfixes and performance improvements"
So yes, they did say the performance should be better.
Stellaris Performance Analysis #2
- Thread starter serpentskirt
- Start date
-
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.
Also the 2.2 patchnotes mention:
"Numerous other bugfixes and performance improvements"
So yes, they did say the performance should be better.
The only solution isn't reducing pop numbers, optimizing the code for them can help a lot as well.
There are other options as well, like spacing out different calculations for them. For example do we need to check jobs for every pop everyday? What if everyday checked 1/5th of pops instead, so over a 5 day period every pop would be checked, add in a few circumstances where every pop is re-checked like on planet conquest, though would only have to be for the two empires involved in the changes, or we could even be fine with having to wait the 5 days. This could gives som problems and bugs as well, and is just meant as an example.
Another example could be, what if we off set every empires calendar by a single day? Wouldn't help on the day-to-day calculations but could reduce the monthly tick over delay.
There are other options as well, like spacing out different calculations for them. For example do we need to check jobs for every pop everyday? What if everyday checked 1/5th of pops instead, so over a 5 day period every pop would be checked, add in a few circumstances where every pop is re-checked like on planet conquest, though would only have to be for the two empires involved in the changes, or we could even be fine with having to wait the 5 days. This could gives som problems and bugs as well, and is just meant as an example.
Another example could be, what if we off set every empires calendar by a single day? Wouldn't help on the day-to-day calculations but could reduce the monthly tick over delay.
Or just remove pop “checks” entirely and let the user specify what pops prioritize what jobs. With a manually activated “auto calc best” that could be toggled if they wanted. Have the AI only use it every couple years.
Yeah there are plenty of ways to optimize...but at its root it’s more pops checking more things. Previously i don’t think they checked anything... and there were less of them. If this is the case I don’t see how this system can ever match the old performance wise.
Yeah there are plenty of ways to optimize...but at its root it’s more pops checking more things. Previously i don’t think they checked anything... and there were less of them. If this is the case I don’t see how this system can ever match the old performance wise.
There's a mod that removes "trade range" from trading hubs and adds a "trade range" of 1 to any outpost, thus eliminating the fact that the game supposedly reacalculates trade routes every day or month. People using that mod said that they did get a performance boost, albeit it wasn't the be-all-win-all solution. Has this been looked into?
Seeing how The Great Strategy Game chokes shuffling 10K pops' icons between slots, for some reason, I recall a small $5 indie game called Banished, made by just one guy. It also chokes at around the same number of population, tbh. But those 10K villagers actually simulated - dozens of their dynamic life conditions, hand storage, pathing, duties they perform, animation and so on.
Think about this comparison, people.
Think about this comparison, people.
I have no idea how inefficient the game engine should be to choke like that with 10K pops. Do everyday ticks all scripted and interpreted in runtime or what?!
Seeing how The Great Strategy Game chokes shuffling 10K pops' icons between slots, for some reason, I recall a small $5 indie game called Banished, made by just one guy. It also chokes at around the same number of population, tbh. But those 10K villagers actually simulated - dozens of their dynamic life conditions, hand storage, pathing, duties they perform, animation and so on.
Think about this comparison, people.
Banished is completely written on C++ as far as I remember.
If you want a better example of effectiveness check KolibriOS. An OS written on clear Assembly language. It is perfect in terms of effectiveness. Not a game, though.
I wonder if anyone has done tests on the absolute mimimum pc to play this game. I would imagine its so bad that it might be impossible to play the game at this point on specific galaxy settings. Makes you wonder if its possible to take legal action if a product is not working as advertised provided you have the necessary prerequisites.
I'm really starting to wonder is there is a much bigger problem with the game-engine itself, rather then just some badly thought calculation principles. And that saddens me the most. It's one thing to have a bit of a bad code, it's another things to find out, 2 years down the line into game release and active support, that entire game engine is actually not fit to your game.
I still don't get how Stellaris needs more than 1 second to save 3 MB (Year: 2371) (autosave) and freezes the game UI until it's done. You need to compress the save file? Here, let me tell you about background threads! You need to gather additional information? Well, it's 3MB of information to save, much less need to calculate all that, if calculation is needed. Considering what a computer can do in 1s, it should be a piece-of-cake to:
1. gather all necessary data
2. create a background thread
3. do necessary calculations on background thread
4. save data to HDD
And even if it takes a bit longer: Doing this in background threads would solve the problem of having the UI freeze! First rule of UI programming: You don't freeze the UI.
1. gather all necessary data
2. create a background thread
3. do necessary calculations on background thread
4. save data to HDD
And even if it takes a bit longer: Doing this in background threads would solve the problem of having the UI freeze! First rule of UI programming: You don't freeze the UI.
A bold claim then, indeed.
Don't forget that the need of making stuff desync-safe may contribute to some performance sacrifices (that's probably why 2.1.4 is slower than 2.1.2).
Gestalt only with no wormholes/gateways session has ended in spectacular CTD (that several users have already reported) and is a perfect example why statements should be based on multiple samples. For starters, as number of recorded games increased more than twice, let's recalculate mean data:
And pruned 200 years in:
So, same setup, but without wormholes and doesn't look better at all, if barely equal. Why? Correct answer - not enough information.
Guesstimation before dissecting savegame files - a lot more hostilities were going on. How so? I actually lied and created 11 gestalts (6 hives and 5 machines) as I was not sure if 2 advanced AIs do count towards overall number (they do not). So could've been that this session had devouring swarm while previous hadn't. Slim chances, but you know that a million-to-one chance succeeds nine times out of ten.
Second possibility - in previous setup some hives conquered robots (or vice versa) and quite an extinction occurred in 2280s thus improving performance. As stated above - not enough information.
I'll abstain from making further conclusions before extracting additional numbers.
I don’t know what data you need to make that graph (if you need to run some background program on your computer or your parsing stuff directly into a spreadsheet) but i’d be willing to run some simulations too for ya if that’s workable and you have a base set of settings to work from. Although adding a different machine into the mix probably isn’t as useful now that i think about it. More noise.
I booted up and old save that was from 2.1 mid crisis (loaded into 2.2.3 now so trade routes are added) to see if a steam mod fixed Scourge behavior (ie let them purge). Large galaxy, stil many empires remaining. But very few pops and tiles as they were hold overs from the old system. Trade routes were dropped in but on the whole it was far more stable than anything i’ve seen yet that late in the game. Too bad there isn’t some easy parser to just chew through the saves and pull out things like
1. number of planets inhabited
2. number of pops.
3. number of districts/buildings etc
etc. for some controls...it would add a bunch of color to that graph.
I booted up and old save that was from 2.1 mid crisis (loaded into 2.2.3 now so trade routes are added) to see if a steam mod fixed Scourge behavior (ie let them purge). Large galaxy, stil many empires remaining. But very few pops and tiles as they were hold overs from the old system. Trade routes were dropped in but on the whole it was far more stable than anything i’ve seen yet that late in the game. Too bad there isn’t some easy parser to just chew through the saves and pull out things like
1. number of planets inhabited
2. number of pops.
3. number of districts/buildings etc
etc. for some controls...it would add a bunch of color to that graph.
Last edited:
Both of those seem like they could easily be reduced to once-per-month tasks, if even that. It could probably be reduced to even once or twice a year per pop by breaking them into the necessary groups by some criteria or another. And the trade route updates could probably be something like every three months and they'd still work fine.
If they're really being checked every day, that's absolute madness, and madness they have multiple other games of experience with to tell them why it's madness at that (eg. EU4 used to update various trade calculations on a daily basis, which used unnecessary calculation time and led to obnoxious stuff like losing and gaining bonuses for trading in certain things every few days).
For those wondering about Glavius AI and pop job checking, here's a bit of his update log on January 2nd :
I don't have any figures, and I didn't get past 2300 with this as I got surprised and crushed, but my subjective impression is that even so early it's night and day. Would be nice to run these tests with his mod.
I don't have any figures, and I didn't get past 2300 with this as I got surprised and crushed, but my subjective impression is that even so early it's night and day. Would be nice to run these tests with his mod.
I've been running all my games with his mod active. My experience is that I don't see any improvement in tick "performance" per say (ie speed of the game as time goes by) but at the same time the AI is no longer brain dead retarded. So he's managed to tinker out 100x better AI performance for no additional computational cost. But the root cause of the cost is still there.
My fear is that (in addition to other extraneous stuff like trade bugs we've all heard about) its the sheer number of pops/tiles themselves. I've loaded up some 2.1 games into 2.2.3 where all empires are still present in 2450+ in a huge galaxy but the number of pops is 10-100 times lower than what you might see in a 2.2.3 game and i'm getting 2x performance off the bat. Even with trade routes added in ticking with whatever jacked up pathfinding they have. That boosts me to about 20-40 fps which is at least playable.
I never play without either, and prior to this specific update from him, perf was worse than vanilla actually. I only played vanilla 2.2.3 once so I honestly can't make a real comparison. Just it seems better to me from what I remember. Hence why I think it's worth a shot to make at least a test run with the mod taking one of the previously tested settings to compare and get objective results.
Anyway if pop jobs are checked daily as claimed earlier in this thread, number of useless operations has to be checked before number of pops. Who knows what other thing that is inadapted to the new model the code is doing?
Regarding loading a 2.1 save into 2.2.3, does that really work?
Not calling you a liar, but if you get better perf for the cogs being loose, I wouldn't say that's a valid point. 
I never play without either, and prior to this specific update from him, perf was worse than vanilla actually. I only played vanilla 2.2.3 once so I honestly can't make a real comparison. Just it seems better to me from what I remember. Hence why I think it's worth a shot to make at least a test run with the mod taking one of the previously tested settings to compare and get objective results.
Anyway if pop jobs are checked daily as claimed earlier in this thread, number of useless operations has to be checked before number of pops. Who knows what other thing that is inadapted to the new model the code is doing?
Regarding loading a 2.1 save into 2.2.3, does that really work?
I was as surprised as you..it actually does work. The only reason i tried it was b/c it was the only game i had that had crisis spawned and wanted to see if a mod fixed it.
It’s aint perfect...all planets have only city districts for example...and the pop counts/0 buildings present on all planets make any economy the AI has useless. But it loaded and ran.