Kernels past 5.3 have a huge performance downgrade in emulators and some games

I don´t know how the people who can fix this could notice this problem (I sure can´t so I ask you all to do it if you know how), but in newer kernels somethings are not working as they should.
I had exactly the same problems listed in this post:
(I can´t paste links for some reason)
But I have an Intel CPU, so this problem is not exclusive to that AMD CPU.
And I don´t see any improvement to this day, so if any developer is reading this, please fix this bug or contact whoever could do it to fix it, it´s absolutely impossible to play certain emulators/games in Linux right now (mostly in Arch based distros according to the person that redacted the post in the Archlinux forums).

you have a haswell intel family ,
also ivy is concerned for kernel 5.4+
i suspect microcode for these family as you may not have update bios for vulnerabilities ( 2018+ )

Here is some information on a possible problem in the Kernel. On the first page is an excellent response from Linus Torvalds:

And what is wrong with the haswell family? I didn´t understand that.
Who is ivy? sorry that I don´t know.
The last update for my bios is from 2016 so I can´t update it.

Can you give some hardware info?
inxi -Fza

Does that mean that this problem is just a part of a bigger problem in recent Linux kernels? And they are trying to fix it already?

I have a small Audio PC with Intel Celeron N3150 CPU.
Runs fine with kernel 4.19.

The basic problem is integral to the concept of a “scheduler”. IMHO, I think as software had become more complicate and hardware has now added multiple cores to the CPU, scheduling is going to get more complicated. At least, that is my opinion. Read carefully what Linus writes.

about ivy bridge , haswell , this was time just before skylake appears ,
and some manufacturers motherboards was started to cheat on turbo pstates frequencies that go beyond limit cpu , Intel change to not take anymore acpi frequencies turbo for that , and it was called hwp.

in fact intel has to change this before skylake appears , and after some update bios , you can have in this case soft pstate but no hwp state, also some laptops are also concerned if there is no hwp pstate

see more on this

1 Like

As I said, the problem is in kernels post 5.3, 4.19 works fine for me too.

I understand that, but my problem, as said in the post of the archlinux forum, is only present in archlinux distros, so it´s not impossible to do it right, it´s just a problem with the archlinux kernels or some packages related to it.

what would you recommend me to do to solve this? Should I disable pstate or cstate from bios?

@RBNMN Check out my post above for the research I did when I started having performance issues with the most recent stable kernel. You can ignore the part regarding the kernel parameter if you didn’t make any changes to grub. Good luck and Cheers!

1 Like

Hi, ty so much for telling me this. I will check when 5.9 becomes more stable and try it and see if it´s solved already, but if not, I´ll try your workaround.

1 Like

Our computers are very different so I haven’t looked into how stable kernel 5.9 is with an NVIDIA GPU, but I’ve had no issues with my Intel based Dell laptop. Your system is very similar to my Windows 10 desktop though funny enough, but I haven’t tried Manjaro on it, someday maybe.

Kernel 5.10 will be the next LTS release to reach Stable in December, after the latest LTS kernel 5.4 so that may work out for you if you decide to upgrade the kernel at that time (assuming you prefer LTS kernels) and update the CPU governor. Good luck!

1 Like

Yes, I will use the new 5.10 kernel if it´s LTS, then I´ll see how everything works, maybe I´ll update this post then.
Thanks for the info and have a good day!

1 Like

After today’s update, even with Kernel 5.4 Citra runs fine, it went from 45-55 fps to 100-130 when fps unlocked.
I don´t know if the solution is related to the nvidia drivers (I had to update them in this new Manjaro version because I had some problems with the old one, which was 430xx and now I have 455 prime) or related to some minor update of the lts kernel 5.4, but it works fine. I didn´t try kernel 5.10 yet, but probably will be fine too.

So between this anecdote and your earlier inxi etc …
I guess I will conclude you were mismanaging your drivers.
It appears you not only had old nvidia modules … but you also have a hybrid/optimus/dual-gpu machine … meaning you should have been using PRIME … which apparently you werent.
(not to mention only nvidia (455) and nvidia390 are supported)

So … yeah. broken driver modules, probably related to not following proper procedure for installing and configuring drivers (IE - using mhwd not pacman directly):

Now that you have a current (not EOL) kernel, and the correct drivers, it works… not very surprising.