I really need to set up my older machine and test it too. The support desk has seen some issues with newer hardware that only happens with newer hardware...
Performance since the patch has rendered the game unplayable
- Thread starter Grim Konstantin
- 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.
depends on how you phrase "should". Almost nothing runs in parallelism perfectly except those things that can run without being tied to each other. Off loading to parallelism also incurs diminishing returns.
An example
The store can handle 100 dozen cookies on the shelf, and stocks it once a day
baker's only have one cookie sheet, it takes 5 minutes minutes to prepare the cookie dough and put it on the sheet. the cookies take 10 minutes to cook. the oven can only handle 1 sheet at a time, and every 10 minutes they get a 5 minute break. Also union rules say they work no more than 6 hrs a day.
How many bakers and cookie sheets each should they have for maximum profit?
Hello, here's a few interesting facts:
1) The game clearly gets CPU limited - overclocking single core raises the date by some. This is on Threadripper 1950x. I checked 3.4 through 4.2 GHz. Gets about 50% further until choking. (Defined as taking 10s per day on fastest.)
2) Drivers do not matter at least on Vega FE. Tested from 17.9 up to latest 18.12.1. Including Pro mode. Resolution does not matter either.
3) Issue seems to be related to jump drive prevalence, but I'm not 100% sure about that, need to experiment with this. Apparently some bit and bob of code in pathfinding is *really* slow when jump drives exist. This is why there are commonly steps at ~2300, 2350 and 2400 which are typical dates for many AI empires to get to jump drives. This effect started to happen since 2.0 but increased in magnitude in 2.1 and 2.2. AI seems to be doing something stupid with jump option? Issue gets way worse when an empire with jump drives has access to gateways or l-gates.
4) There's also some exponential slowness code related to number of empires each AI empire has in diplomacy range. The game is much faster if you force everyone to be xenophobe. Megacorps are the slowest followed by federations.
As for AI and pathfinding can be done at RTS speeds for many thousands of units nowadays on a single core! I remember ancient RTS from 1996 running 1000 units simulatenously on far worse PCs. Heck, something like Supreme Commander which was a bit newer... Stellaris does not even come close. It's not about optimization, it's about bad algorithms...
1) The game clearly gets CPU limited - overclocking single core raises the date by some. This is on Threadripper 1950x. I checked 3.4 through 4.2 GHz. Gets about 50% further until choking. (Defined as taking 10s per day on fastest.)
2) Drivers do not matter at least on Vega FE. Tested from 17.9 up to latest 18.12.1. Including Pro mode. Resolution does not matter either.
3) Issue seems to be related to jump drive prevalence, but I'm not 100% sure about that, need to experiment with this. Apparently some bit and bob of code in pathfinding is *really* slow when jump drives exist. This is why there are commonly steps at ~2300, 2350 and 2400 which are typical dates for many AI empires to get to jump drives. This effect started to happen since 2.0 but increased in magnitude in 2.1 and 2.2. AI seems to be doing something stupid with jump option? Issue gets way worse when an empire with jump drives has access to gateways or l-gates.
4) There's also some exponential slowness code related to number of empires each AI empire has in diplomacy range. The game is much faster if you force everyone to be xenophobe. Megacorps are the slowest followed by federations.
As for AI and pathfinding can be done at RTS speeds for many thousands of units nowadays on a single core! I remember ancient RTS from 1996 running 1000 units simulatenously on far worse PCs. Heck, something like Supreme Commander which was a bit newer... Stellaris does not even come close. It's not about optimization, it's about bad algorithms...
Last edited:
ummm I've have yet to see any month take me > 17 sec much less a day take 10 seconds. Would like a save of this to test.
Allow me to rephrase:
I don't think it was doing that in older versions of the game. Is that kind of load between cores new with the current patch? Or was I just misinformed about older versions of the game?
not really, at least not when I've tested HOI4 and EU4. Haven't given stellaris as much love for testing. PDS games tend to be heaviest on 1 core / hyper thread in my experience, but what I'm seeing here isn't par for the course. Will set up the old 4770 w 650ti and test it on it to see how it fairs as it was the most stable test machine ever for me and PDS games.
Ah, I was looking for a post like this!
I have major camera panning stutter even though on steam overlay the FPS is above 60. The previous version didn't cause this.
Nothing about my hardware or drivers/settings changed from the previous patch to the current one.
I have major camera panning stutter even though on steam overlay the FPS is above 60. The previous version didn't cause this.
Nothing about my hardware or drivers/settings changed from the previous patch to the current one.
Yeah umm... if you cant play halfway through the game it isn't in a playable state. The game is essentially out of commission right now.
Cant get into the 2500's (at 2476 now, took us HOURS to play through 30 years today) on a 1k star map. We've done 3-6 man games regularly in the past and sure it got slow as time went on, but never this stuttering that prevents you from playing more than half a game before becoming unplayable. We've planet popped over half the galaxy to try and kill some lag while we wait for the 5x crisis to pop up but we'll literally never get to end game like this as the lag and stuttering gets progressively worse grinding the game to a halt as it has now.
Only info our small group can contribute to solving the problem...
-Planet popping to reduce things going on and removing most of the AI doesn't seem to help the situation.
-The 3 of us have millions in fleet power and repeatable tech so, number of ships or a memory leak isn't out of the question
-2 of us are building quiet tall, I run 220 pops on ringworlds with tons of factions going on because I poorly managed them so.. lots of different factions,species, and pops on a single entity.
That's all I've got .__.
Cant get into the 2500's (at 2476 now, took us HOURS to play through 30 years today) on a 1k star map. We've done 3-6 man games regularly in the past and sure it got slow as time went on, but never this stuttering that prevents you from playing more than half a game before becoming unplayable. We've planet popped over half the galaxy to try and kill some lag while we wait for the 5x crisis to pop up but we'll literally never get to end game like this as the lag and stuttering gets progressively worse grinding the game to a halt as it has now.
Only info our small group can contribute to solving the problem...
-Planet popping to reduce things going on and removing most of the AI doesn't seem to help the situation.
-The 3 of us have millions in fleet power and repeatable tech so, number of ships or a memory leak isn't out of the question
-2 of us are building quiet tall, I run 220 pops on ringworlds with tons of factions going on because I poorly managed them so.. lots of different factions,species, and pops on a single entity.
That's all I've got .__.
I played another game (this time on small galaxy, contrary to Large I usually play). With default numbers of empires for small galaxy. And curious thing - game still slows down at around the same time. Slowdown is less, due to smaller size of, well, everything, but it's still happening very sharply 100 years into the game and still very noticeably big. Month on fastest already taking 12 seconds - which is better when played on Large (around 16 seconds around this game-time), but still a 2x slowdown compared to game start. So 100% there is something wonky with the code - if even reducing the size and number of actors doesn't make that much of difference.. I'm curious what game would behave like if you start a 600 star galaxy with just 2 empires, no FE, no primitives and they would basically share the galaxy between the two.
I really, really hope you're right.
My big fear here is that the stuttering is due to an inherent flaw in the Clausewitz engine. Stellaris has had slight stuttering problems right from the start and all the other Clausewitz games have them too ( though a lot less pronounced and, for me, very tolerable ). It just has never been as bad as it is now. There are threads on this topic both on Steam and on these forums going back all the way to the game's release in 2016.
The devs seem to be very much aware of the issue, too. Here's a quote by Guraan from another thread: "Yeah the daily tick stutter has always been a issue, but now i agree its ridiculous (even on normal speed)." ( https://forum.paradoxplaza.com/foru...tering-skipping-with-every-game-tick.1133794/ ).
The fact that they have known about the problem all this time and it hasn't been fixed yet, doesn't exactly fill me with confidence...
I don't know if you've been following it, but this thread is also a very good read on the topic: https://forum.paradoxplaza.com/foru...tellaris-a-quick-performance-analysis.1138327
Stellaris is slated to be released on consoles in 2019, right? Hopefully the devs get a grip on the performance problems until then.
If the game hits the shelves with these issues unsolved, Paradox will be in a world of hurt, I fear.
You need to put max settings. Large galaxy, high number of civs, advanced_galaxy debug thing to kickstart it, then let it roll some more years ahead. If you're lazy, also use human_ai debug command. Advanced AI starts also helps with reproduction. It really does get that slow by some 2450 equivalent. (advanced_galaxy itself does not build enough things to slow it down)
In typical playthrough, I get 2s per day in very late game large galaxies.
For stuttering to occur two simultaneous things has to be true:
1. Rendering thread waiting on game loop, which happens in Stellaris, because rendering thread is not separated from thread handling daily ticks (https://forum.paradoxplaza.com/foru...ellaris-a-quick-performance-analysis.1138327/)
- I'm not sure if this is property of Clausewitz or just property of Stellaris, only devs will know this.
- either way fixing this would require major rewrite, which is reason why any optimization will likely be in 2.
2. Amount of work that game thread has to do between some two frames takes more time to calculate then frame time
- so game needs to evenly spread work of daily tick between multiple frames and dynamically slow down tick speed and spread work between more frames when of amount of work is too high
Multi-threading of work of-course helps too, but it's not inherently sufficient to prevent stutter, spreading of work is more important. More multi-threading should however increase overall tick speed as long as game will spread work between less frames.
My guess is, that devs put calculations of some new system just between two frames and now they have to either reduce amount of calculations in that system or find the way to divide that work between multiple frames. This is often not trivial task and requires time.
I agree 100%.
I've made similar comments regarding the spreading of the workload in the thread I quoted earlier.
For me, Glavius AI mod has now fixed the stuttering problems. He did some changes to the daily pop job calculations and pathfinding and now I'm able to play beyond 2500 without much problems. The game is running now like in 2.1.
So I was also encountering severe performance problems in my playthrough which became really noticeable around 2300 and got signifiantly worse as time went on. I just switched to the beta patch to see if anything would improve.
Here are the changes: before I switched, I couldn't even run it on slowest without daily stutter and everytime i clicked anything in any menu the game would just freeze for a few seconds. These problems are all gone, but now it freezes every 10 days or so for about 7 seconds regardless of speed setting.
The save is currently in 2390, so a bit before the afore mentioned 2400 mark.
My specs are:
Intel Core i3-6300T CPU @ 3.30GHz ~3.30 GHz
8 GB RAM
Nvidia Geforce GTX 1070
Something I also observed is that one of the cores on my CPU is always at 100%, while in regular intervals (presumably about 10 in-game days) the other cores seem to catch up only for the game to freeze and them dropping to ~0% again.
The save I attached is the one from the beta patch.
Here are the changes: before I switched, I couldn't even run it on slowest without daily stutter and everytime i clicked anything in any menu the game would just freeze for a few seconds. These problems are all gone, but now it freezes every 10 days or so for about 7 seconds regardless of speed setting.
The save is currently in 2390, so a bit before the afore mentioned 2400 mark.
My specs are:
Intel Core i3-6300T CPU @ 3.30GHz ~3.30 GHz
8 GB RAM
Nvidia Geforce GTX 1070
Something I also observed is that one of the cores on my CPU is always at 100%, while in regular intervals (presumably about 10 in-game days) the other cores seem to catch up only for the game to freeze and them dropping to ~0% again.
The save I attached is the one from the beta patch.
Concur. I play with random numbers of AIs, and games with more extant AIs are far slower than the others.
Well I also started another game on a 600 star galaxy, after I finally gave up on finishing my previous game on an 1000 star galaxy due to the game's abysmally bad performance. I was shocked to see that it didn't really seem to matter. The game again has slowed down to an unbearable degree. Soon I'll have to give up that game also. I'm wondering now if even a small galaxy will do that if you play a long enough game. And if galaxy size is not a primary factor, what is?
Im not saying you are making it up I specifically took care to say maybe some people didn't notice performance drops.
As for why you should do it, you obviously don't. But since neither you nor do any of the people who claim to have no performance issues provide no proof whatsoever, as oposed to the people who actually say they have performance issues, my opinion stands.