Bad Performance (Cornflake 1.5)

[Corner79]

one was released as beta earlier today

Sweet! I didn't see it yet, I have been away from home since Friday morning.

 

[Corner79]

Oh completely agree! Why I go out of my way to mention that I don't actually know if this is the intended performance, still not above it being say a driver issue or whatever. Simply argued a hypothetical

And that's why this forum is the best among so many, it's a true gaming community, we are all in this fun and history together

 

[Night Light]

And that's why this forum is the best among so many, it's a true gaming community, we are all in this fun and history together

But nothing in changelog about our problem. Nothing interesting.

 

[Xiziz]

Its late, so I'll just drop this table here. @podcat

Oh, still no DLC, just patch, should have been in the table.

All differences except opengl can probably be attributed to when Spanish civil war kicks off (later SCW start = faster 1st year)

 

[arithon]

Haven't played the new dlc or patch yet, tonight.

Can anyone say if the division spam has lessened to any extent ? That has always seemed to be the culprit to me!

 

[Night Light]

Haven't played the new dlc or patch yet, tonight.

Can anyone say if the division spam has lessened to any extent ? That has always seemed to be the culprit to me!

No. In my opinion ,AI started to make much more divisions than on 1.4.2.

 

[BurdenedWarrior]

Haven't played the new dlc or patch yet, tonight.

Can anyone say if the division spam has lessened to any extent ? That has always seemed to be the culprit to me!

patch log for 1.5 mentioned that AI builds a lot less units(perhaps 1/2 of 1.4.2), I haven't played enough to notice though.

Its late, so I'll just drop this table here. @podcat

Oh, still no DLC, just patch, should have been in the table.

All differences except opengl can probably be attributed to when Spanish civil war kicks off (later SCW start = faster 1st year)

Nice work @Xiziz, good to see some hard data on changes. I wish we could do a test for 1942 to 43. But I don't think it is possible to negate the variables to set up a comparable test.

 

[Limith]

I run Hearts of Iron from my E: Drive, steam and everything aswell, and tend to have 40 tabs, a game and a couple other things open all at the same time, it's a mess.

Regardless, this was not uncommon before the update, it remains that the issue occurred after the update and that there has been no similar slowdown in other Paradox titles or the like. I'm not complaining about slow speed, i'm complaining about the DRAMATIC slowdown in 36 as to before the update, it ran by seamlessly at ever 24 hours, now it takes a good bit.

Close your browser tabs, that's 40 threads which are running in the background, and who knows what the javascript ads are doing. Since HoI4 now uses parallelization (see my previous explanation of how context switches ruin speed) the game is affected by your background threads significantly more than previous. Tl;DR close everything but HoI4 and Steam and the performance should be significantly better. It doesn't matter if you aren't hitting "100% CPU" usage. That's a false statistic. Just the impact of context switches will slow down you performance tremendously.

For people with below specs and why previous game ran better:
Paradox parallelized the game to use 4 cores. If you have 4 cores, this makes your game faster. If you do NOT have 4 cores, this makes your game slower, as your OS needs to do 'context switches'.

Analogy: assembling a car

You have 4 workers and 4 tires to put on a car.

Previously in serial:
1 worker puts on a tire at a time, and can immediately put on the next tire without traveling to the next location (say the worker stands in the middle of the frame.

Now in parallel:
Each worker stands next to where a tire will be and puts them on at the same time. You are now 4x faster.

HOWEVER if you only have 2 workers:
Each worker puts 2 tires, then has to 'travel' (context switch) to putting on the other two tires, adding time and slowing down the process. Due to how computers work, the workers have to travel between the two locations multiple times. What happens is the workers putting on a bolt on one tire, then the worker walks (context switches) to the other tire, put on another bolt, walk back, put on a third bolt, etc etc. So the constant switches slows down your thoroughput.

___
It doesn't matter if your CPU is or is not maxed out before or after with the patch. Context switches by itself slows down the game. In addition, parallelization usually requires more complex code logic that also has a performance impact (which should be offset if you actually have more cores but in this case you don't).

In this case, you are having 4 workers constantly switching between 40+ threads (tasks) which will dramatically worsen performance.

(Also I know that some version of Chrome has a bug where it will use up 1-2 cores worth of CPU randomly, and was causing my work laptop to overheat when I had basically only 4 tabs open).

Most likely the background AI threads aren't performing as fast on your rig and at 24H there is a synchronization wait (the main thread waits for the AI thread to finish).
____
@Had a dad The gist of it is even though people may be meeting minimum specs 'hardware wise' the usage of their machines means they aren't actually meeting minimum specs as they don't have the number of free threads required. You should mention in the how to provide stats area to close all background processes.
_____
With regards to the dramatic slowdown in 36, I read the technical dev diary and I believe the system learns what the average tick size is and allocates time for it. It may be the learning algorithm has not had enough time to optimize for a good tick size OR the additional early game setup leads to unexpectedly large tick sizes. Just a guess. This is a parameter tuning problem that requires some trial and error to specify a good starting tick amount. It may be better to just hardcode it (or softcode it) rather than rely on algorithmic calculation. Eg similar to how I learned that the total war games have implemented more and more complex reinforcement learning algorithms when the top 2/3 rated best TW games didn't have complex ai algorithms but a simple 220 list of if else statements (and their initial foray into reinforcement learning Rome 2 had terrible AI) Sometimes the 'geometric' solution performs better than the algorithmic one (given performance constraints)

 

[Emma354]

Close your browser tabs, that's 40 threads which are running in the background, and who knows what the javascript ads are doing. Since HoI4 now uses parallelization (see my previous explanation of how context switches ruin speed) the game is affected by your background threads significantly more than previous. Tl;DR close everything but HoI4 and Steam and the performance should be significantly better. It doesn't matter if you aren't hitting "100% CPU" usage. That's a false statistic. Just the impact of context switches will slow down you performance tremendously.

No offence, but i'm smart enough to have tried something that obvious.

 

[steve213]

the game just crashes everytime I try to run

 

[Corner79]

the game just crashes everytime I try to run

Report to the bug forum, so they can properly help you.

This post is about slow performance issues

 

[Cagarustus]

So, if people are having issues with performance after 1.5, this begs the question: did paradox actually fix it?

 

[Corner79]

So, if people are having issues with performance after 1.5, this begs the question: did paradox actually fix it?

Late game lag is fixed, that 40-onward lag. So that's a plus!

There is some new issues is what this thread goes into. Some experience a slow 36-37, some a noticeable lag at the 24 hour rotation, and a possible sound file problem causing screen lag when scrolling over combat. That's just a quick summary

 

[Sun_Wu]

Also, I looked at Intel processors and to change the motherboard to get intel would be more money that I am willing to spend.

What I am looking at is getting this motherboard https://www.newegg.com/Product/Product.aspx?Item=N82E16813132988
and get this processor https://www.newegg.com/Product/Product.aspx?Item=N82E16819113434

That is an excellent CPU that should work for the rest of HOI4's development

 

[Tavior]

I'm biased against AMD as over the last 6 yrs I've built 5 PC's with only 1 being an AMD, guess which one I have issues with games on...

That said, yes that cpu is above recommended specs, but please note you most certainly would need a different motherboard with that.

I would have like to point out that my PC is Ryzen 1700X and Vega 64. I don't have any issue with HoI 4 in general.

Although there was an issue regarding the driver that AMD did not fix for a while earlier although. I never had any real problem with AMD PC just because it was AMD components. If anything I had more problem with power supply. My luck seem to be pretty bad with them.

I am favoring AMD pretty heavily recent because Nvidia is dominating the GPU market and we, customers, don't want them to have monopoly. In which case they would just jack up the price as much they wanted to.

 

[Sun_Wu]

I would have like to point out that my PC is Ryzen 1700X and Vega 64. I don't have any issue with HoI 4 in general.

Although there was an issue regarding the driver that AMD did not fix for a while earlier although. I never had any real problem with AMD PC just because it was AMD components. If anything I had more problem with power supply. My luck seem to be pretty bad with them.

I am favoring AMD pretty heavily recent because Nvidia is dominating the GPU market and we, customers, don't want them to have monopoly. In which case they would just jack up the price as much they wanted to.

AMD's CPUs were inferior to Intel in gaming by 2010 and got worse over time until Ryzen launched. Ryzen is really a return to glory for AMD as it harkens back to their magnificent Athlon XP and Athlon 64 days. The Ryzen platform is a little worse on single threaded performance, much better on multithreaded performance and much cheaper. They are solid chips for PDX games.

 

[Tavior]

AMD's CPUs were inferior to Intel in gaming by 2010 and got worse over time until Ryzen launched. Ryzen is really a return to glory for AMD as it harkens back to their magnificent Athlon XP and Athlon 64 days. The Ryzen platform is a little worse on single threaded performance, much better on multithreaded performance and much cheaper. They are solid chips for PDX games.

Indeed I am fine with the slight worse single thread performance because more and more game developers are starting to embrace multi-thread. Who know someday multi-thread performance may become more critical than single thread performance.

 

[Misaka_Complex]

This issue it still present in 1.5.1, how about for others?

 

[Killerrabbit]

This issue it still present in 1.5.1, how about for others?

Same.

For me the beta 1.5.1 update where they added in a small optimization to provinces did improve things slightly, but it is still very much slower than 1.4 - specifically the 24 hour tick lag is pretty bad. The actual speed for normal 1-hour ticks is fast and responsive enough for me.

My CPU is not state of the art at all, but it is a older i7 4-core that performs good enough for any other modern games and resource intensive tasks.

 

[Sanny]

I get lots of stuttering even when paused. I have a GTX 1080 and an i7 4770k, so obviously its the game's optimisation.