The problem is: Plasma performance tanks when using the iGPU with a external 4k monitor.
We are working still working on this, meanwhile can you check if you still facing this issue on latest driver? If possible can you share bug-report as it will gives clarity for creating exact local setup to replicate issue.
Driver Version: 590.48.01
the issue is still here when theres an external monitor connected to the optimus laptop
the external monitor generally gets half of the fps its supposed to get
while the laptop display is normal (120 fps in my case since its 120hz)
kde plasma version is 6.5.5 and im on arch
This is still an issue as of latest driver 590.48.01.
KDE Plasma 6.5 (And 6.6 beta 2) Wayland: KWin running on iGPU, dGPU (Nvidia) display output (generally) runs at half (or less) of display refresh rate. Frame pacing is also questionable.
Running KWin on the Nvidia GPU is a workaround and not a solution.
Please provide us with an update.
Can you please share latest bug report from 590 driver? This will help us to debug this issue on 590 drivers as well.
Can you please share latest bug report from 590 driver? This will help us to debug this issue on 590 drivers as well.
latest driver (i think so) after sudo pacman -syu
the issue still persists im getting 60fps on my external monitor while primary laptop one is 120fps
package version is local/nvidia-open 590.48.01-12
with nvidia smi its the same version 590.48.0
Yes, although I’m not sure how a bug report would help seeing as this is a performance issue and not a stability one.
Additional note: when running KWin on the iGPU and using the Nvidia dGPU display output (so, when the conditions for this issue are met), I am still able to get a full 240Hz output if the GPU has a load on it (for example, vkcube ran on the Nvidia GPU with –present_mode 0 (no frame limit))
Perhaps GPU copies slow down when the card is near idle?
The bug report attached to this reply was taken in the following conditions:
KWin running on AMD Ryzen 4800H iGPU
Ryzen 4800H iGPU driving Laptop Screen
Nvidia GTX 1660Ti Mobile driving External Monitor (Alienware AW2521HFA 240Hz 1080p monitor)
No additional load applied to Nvidia GPU.
nvidia-bug-report.log.gz (529.2 KB)
Yes, still an issue. It’s been almost 20 months.
My Nvidia driver is on Production branch v580.126.09
But seeing fellow users reporting same issue with New Feature branch (590+) as well, I won’t switch
After having used Debian for the longest time I’ve now started using Arch so I could test newer drivers on my dual-GPU T15G G2 laptop with a RTX 3070.
Right now I’m on 590.48.01 and I’m still having issues. I think it has gotten besser but the FPS on the external monitor is still noticably slower (I have quite an old monitor that only goes up to 60Hz and I can still notice it quite a lot).
I think it’s a bit better on Gnome than it’s on KDE but it’s still pretty bad. I’ve switched to dGPU mode in BIOS for now but hopefully the issue will be resolved at some point so I can go back to hybrid mode again.
I’ve got this issue too, cosmic desktop on nixos, driver version 590.48.01. My CPU is an AMD Ryzen AI 350, and I’m on kernel 6.19.2 (I patched your driver to work with the new kernel since it was taking you guys so long). Using a 100hz external monitor and a 120hz internal monitor. Both monitors support VRR and I’ve tried with and without VRR on both.
In my case my 5060ti is in an eGPU enclosure over TB4 so some performance degradation is expected, but not THIS bad. It’s almost as if frames were getting copied over the TB4 cable to the iGPU, and then back to the eGPU over again (at least, that’s what the latency and framerate feels like). But my display is obviously plugged into the eGPU directly so I shouldn’t be getting performance so poor.
I get better performance on my built-in display using an eGPU than I do on an external monitor connected directly to the eGPU. Terrible.
The workaround I’ve been using is to log into a bare gamescope session with the internal display disabled and gamescope running on eGPU only. That fixes the problem entirely, but is not very convenient to have to login to a whole seperate session just to play a game on the GPU I paid full price for.
From what I understand is that, at least on a laptop like mine with a dual GPU set up but the monitor output connected to the dedicated GPU, when you run in hybrid GPU mode everything is rendered on the iGPU. That means that e.g. KDE is rendered completely on the integrated AMD or Intel GPU but for everything to appear on the external monitor it has to be copied onto buffer of the dedicated GPU. That is the point where the slowdown happens. That’s why this issue disappears when you disable the internal GPU. Becuase everything is rendered on the dedicated GPU and no copying happens.
I may be completely wrong, though. I’m not an expert.
Just came to give my monthly update: this is still not fixed in the latest 595 Beta driver.
I can also confirm the same. 595 does not fix this issue. @vrachatte I have provided the bug report you requested (a month ago)
Hello, I’m slowly losing hope this bug is ever going to get fixed. Can we please have an update? Thank you.
This serious problem / bug is still not fixed.
I have an external monitor connected via Thunderbolt cable to my notebook and the performance is terrible (even with flagship GPU inside). The built-in display is OK, no problems there. But everything on the external monitor is managed by my integrated Intel GPU and it’s impossible to change this. MSI doesn’t allow this change even in BIOS (muxless mode).
Wayland + nvidia + external monitor (+ KDE Plasma?) = terrible performance, bad FPS, low refresh rate, buggy and choppy animations.
nvidia please fix this issue, it’s really annoying. It’s been almost 2 years now since this was first reported.
Nope, it’s not specific to KDE Plasma. Same issue with Gnome.
I am quite sure that Nvidia puts all their effort in fixing this bug, even thoug it is already known for years. Otherwise our $4000 GPUs would just be quite expensive bricks and who really wants that? /s
I am sure that the AI Bubble is not the issue here, since it is about Linux and all AI data centres do use Linux. So Nvidia for sure is interested to fix bugs regarding Linux, don’t they? /s
But to be honest, lets support Nvidia by uploading bug reports. Thankfully Nvidia has therefore a genius script, that creates these bug reports.
Use the following script:
nvidia-bug-report.sh
new driver update, new message from me: same issue is present on 595.58.03.
same system configuration as when I took the bug report requested by @vrachatte .
This bug still exists in the latest driver version sadly :(
Bug report:
nvidia-bug-report.log.gz (318.2 KB)