Hi everyone, I’m currently using KDE Plasma 6.5.4 on Archlinux, using the AMDGPU open-source driver (graphics card is Radeon 680M).
I’m almost certain I’m experiencing a video memory leak. Today is the fifth day since my OS booted, and nvtop shows plasmashell’s video memory usage is 2119MB, which doesn’t seem normal. Since I’m using shared memory (4GB total), if plasmashell’s memory usage continues to increase, based on my experience, I’ll encounter frame drops or even crashes in 3D games within two days.
Because this isn’t the first time I’ve encountered this problem, I started recording plasmashell’s daily memory usage five days ago: 290MB on the first day, followed by 519MB, 1684MB, 2021MB, and currently at 2119MB.
I’ve tried to diagnose the problem myself, but unfortunately, there seems to be a lack of diagnostic tools for video memory leaks (or perhaps I missed something). Therefore, I’m seeking help here. If you have any insights, please let me know. Thank you!
I’m sorry I don’t have a solution, but I wanted to comment that I experience the same. Every day seems to increase the VRAM use (and RAM use) until by around day 4 or 5, things are noticeably slower, until I do a reboot. Can’t figure out why, and I also don’t know how to debug this. It’s specifically plasmashell that’s getting higher.
Same issue here with 6.5.4 on Fedora 43.
Workaround is to restart plasmashell: systemctl --user daemon-reload && systemctl restart --user plasma-plasmashell.service
Not sure it would be simple to diagnose this using those.
I am not sure if VRAM is used as regular RAM: in Linux system memory is used as much as possible at any time, but a large part of it is available, only used temporarily opportunistically, so that any application can reuse that memory part as soon as it needs it, but if the data in available memory is accessed again this is instantaneous instead of accessing the disk, improving performance. (https://www.linuxatemyram.com/)
I just moved to 6.5.4 last few weeks from 6.4.5, and I’ve had a lot of weirdness, including this creeping vram issue. It usually starts at a couple hundred meg for both, but quickly grow to linger around this after a few weeks.
This laptop has a 3080rtx with 16gb vram so it’s not killing me, but kwin_wayland and plasmashell both have been using significantly more ram/vram lately in 6.5.
Well one thing I can say for certain: not all hardware are concerned.
I don’t observe this on my intel laptop with itengrated graphics:
$ kinfo
Operating System: Arch Linux
KDE Plasma Version: 6.5.5
KDE Frameworks Version: 6.22.0
Qt Version: 6.10.1
Kernel Version: 6.18.2-arch2-1 (64-bit)
Graphics Platform: Wayland
Processors: 12 × 13th Gen Intel® Core™ i7-1365U
So far the issue has been raised for an Nvidia 3080 Ti and a Radeon 680M, with slightly different symptoms.
Both of those don’t have the best drivers unfortunately, one is proprietary, the other one hasn’t the best stability track record.
This could be driver bugs or some incompatibility with KWin.
It would be interesting to know if there are only those or other GPU/API concerned.
And if this affects other DE that would at least partly tell us whether the issue is in Plasma or in the driver.
My desktop is usually 3x 4k, so 11520x2160 framebuffer, so I expect mine is probably heavier than most others, and has always managed to be a bit wonky around ram/vram use, but it’s definitely higher than I’ve seen it when on 6.4 prior and each 6.x before that. I’ve been watching mine since a last restart a few weeks ago, and it’ll start usually ~500mb, usually hang out around 1.5-2g ram (vram is usually much lower), but this has been steadily climbing.
Of course this could always be nvidia, their driver quality differs greatly and usually playing russian roulette with driver upgrades already with them.
Here’s just a few days after the original, hitting over 5gb of ram and 2.7gb vram. This is about the highest I’ve seen it since upgrading arch last.
That’s just proper memory management, unused ram is wasted ram And windows has been doing this as well for quite a while now. We can blam “RAM cleaner” snake oil peddlers for misconceptions about ram.
Anyhow this is not the case here according to my tests. This memory does not get freed when needed and results in stutter on games with high VRAM usage. On my particular tests normally my workstation sits at about 1.2-1.5G. If I run Dune Awakenings as a repro, total usage gets to 8.5-9G. When this happens, and it’s always after a few of days of uptime, like now:
[$]> nvidia-smi
Wed Mar 11 21:56:19 2026
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 595.45.04 Driver Version: 595.45.04 CUDA Version: 13.2 |
+-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA GeForce RTX 3080 Off | 00000000:65:00.0 On | N/A |
| 0% 46C P8 48W / 370W | 4102MiB / 10240MiB | 11% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| 0 N/A N/A 3240 G /usr/bin/kwin_wayland 130MiB |
| 0 N/A N/A 3337 G /usr/bin/Xwayland 9MiB |
| 0 N/A N/A 3685 G /usr/bin/nheko 43MiB |
| 0 N/A N/A 3713 G /usr/bin/xwaylandvideobridge 3MiB |
| 0 N/A N/A 3716 G /usr/bin/Telegram 53MiB |
| 0 N/A N/A 6370 G ...share/Steam/ubuntu12_32/steam 6MiB |
| 0 N/A N/A 6434 G /usr/lib64/firefox/firefox 216MiB |
| 0 N/A N/A 7599 G ./steamwebhelper 24MiB |
| 0 N/A N/A 7629 C+G ...am/ubuntu12_64/steamwebhelper 8MiB |
| 0 N/A N/A 8258 G C:\windows\system32\explorer.exe 3MiB |
| 0 N/A N/A 993911 G /usr/bin/krunner 10MiB |
| 0 N/A N/A 1642120 G .../Toolbox/apps/clion/bin/clion 27MiB |
| 0 N/A N/A 3766345 G /usr/bin/plasmashell 3030MiB |
+-----------------------------------------------------------------------------------------+
If I run DA now, the game maxes at 10G and starts thrashing. plasmashell usage doesn’t drop. Also this has been hapening for quite some time now and with a lot of different drivers it’s just that lately it has become a lot more aggressive. It’s used to take ~5-7 days to trigger, for the past couple of months or so it happens in 1-2 days of uptime. Also no other application utilizing VRAM exhibits this behavior at all. This is isolated to plasmashell (which obviously does not rule out a driver issue by any means possible).
Debugging the graphics stuck is far from my field or expertise but I will give it as shot as lately it’s getting annoying.
One thing to note that I always notice in the morning, I don’t see it happening gradually, I just see the value suddenly bloated on btop which I have running in a transparent konsole session on my second monitor as a wallpaper, when sit on my workstation to work. So it could possibly be related to dual monitors doing DisplayPort deep sleep (hence disconnected and reconnected) as I’ve noticed that when I see it, xwayland apps also exhibit a problem with opening dialogs in the wrong window. But correlation != causation.
Ok so I gave it shot. There’s nothing exposed in the kerrnel or NVML that can help debug this.
There do not seem to any dma-buf allocations for plasmashell so whatever happens is definitely in driver memory.
The only ways I found out that could help gather more info, are:
RenderDoc, Nsight Graphics and apitrace. All three though do not support attaching to a running process, they need the process launched by them so they can do injection/instrumentation. So this is as far as I can go because I can not run plasmashell on my workstation for days under any of these, and using a VM is not a viable option for the obvious reasons.
Unless someone knows of another way I could collect info?
4060, 4070, 3060, 3070 Ti, 1660 Ti, even 1050 Ti. The only series I haven’t tried yet (bc I don’t have it) is RTX 5000. But it probably won’t be any different because this started happening with Plasma 5.27 and continues even now.
And windows has been doing this as well for quite a while now. We can blam “RAM cleaner” snake oil peddlers for misconceptions about ram.