| Summary: | GPU VRAM sensor reports erratic values | ||
|---|---|---|---|
| Product: | [Applications] plasma-systemmonitor | Reporter: | Bandit <bandit> |
| Component: | general | Assignee: | KSysGuard Developers <ksysguard-bugs> |
| Status: | RESOLVED NOT A BUG | ||
| Severity: | normal | CC: | ahiemstra, asudox+7frg23vcr, kde, nate, plasma-bugs-null |
| Priority: | NOR | ||
| Version First Reported In: | 6.5.0 | ||
| Target Milestone: | --- | ||
| Platform: | Arch Linux | ||
| OS: | Linux | ||
| Latest Commit: | https://invent.kde.org/plasma/ksystemstats/-/commit/fc8120f7500c3bfac1de23b6dc5d53e9eb34de06 | Version Fixed/Implemented In: | 6.5.0 |
| Sentry Crash Report: | |||
| Attachments: |
Sensors shown in 6.5.0 in Systemmonitor
and shown in btop but 1min later |
||
|
Description
Bandit
2025-10-12 07:28:45 UTC
So seems we are wrong by factor 10 approximately? Can you give me your driver version and a few samples of 'nvidia-smi dmon -c 2 -s pucm' thanks ! Hello. Here the requested Info (I hope i did it right):
NVIDIA-SMI 580.95.05 Driver Version: 580.95.05 CUDA Version: 13.0
nvidia-smi dmon -c 2 -s pucm
# gpu pwr gtemp mtemp sm mem enc dec jpg ofa mclk pclk fb bar1 ccpm
# Idx W C C % % % % % % MHz MHz MB MB MB
0 21 56 - 22 27 0 0 0 0 810 450 2594 2630 0
1 16 45 - 0 0 0 0 0 0 405 210 18 19 0
0 21 56 - 26 29 0 0 0 0 405 405 2594 2630 0
1 16 45 - 0 0 0 0 0 0 405 210 18 19 0
and with some load on it:
nvidia-smi dmon -c 2 -s pucm
# gpu pwr gtemp mtemp sm mem enc dec jpg ofa mclk pclk fb bar1 ccpm
# Idx W C C % % % % % % MHz MHz MB MB MB
0 53 64 - 4 3 0 0 0 0 7301 1965 9061 9097 0
1 85 48 - 0 0 0 0 0 0 9251 1725 6468 6468 0
0 46 63 - 4 3 0 0 0 0 7301 1807 9061 9097 0
1 86 48 - 0 0 0 0 0 0 9251 1725 6468 6468 0
On load btop says
GPU0 9,1G / 12G
GPU1 6,7G / 10G
If you need more let me know.
Thank you for looking into it.
It seems like we add fb and bar1 together which results in which means we show ~double the amount in your case. It doesnt explain the tenfold increase that you saw earlier through. A possibly relevant merge request was started @ https://invent.kde.org/plasma/ksystemstats/-/merge_requests/115 Git commit 3c929b0113d356e3242b75a6ef9e0ba962ebe44b by David Redondo. Committed on 14/10/2025 at 08:41. Pushed by davidre into branch 'master'. gpu/nvidia: Only count fb memory This matches what nvidia-settings does to show used memory. In some cases bar can be as high as fb causing wildly incorrect readings. FIXED-IN:6.5 M +3 -3 plugins/gpu/NvidiaSmiProcess.cpp M +3 -3 plugins/gpu/autotests/nvidia.cpp https://invent.kde.org/plasma/ksystemstats/-/commit/3c929b0113d356e3242b75a6ef9e0ba962ebe44b Git commit fc8120f7500c3bfac1de23b6dc5d53e9eb34de06 by David Redondo. Committed on 16/10/2025 at 08:20. Pushed by davidre into branch 'Plasma/6.5'. gpu/nvidia: Only count fb memory This matches what nvidia-settings does to show used memory. In some cases bar can be as high as fb causing wildly incorrect readings. FIXED-IN:6.5 (cherry picked from commit 3c929b0113d356e3242b75a6ef9e0ba962ebe44b) Co-authored-by: David Redondo <kde@david-redondo.de> M +3 -3 plugins/gpu/NvidiaSmiProcess.cpp M +3 -3 plugins/gpu/autotests/nvidia.cpp https://invent.kde.org/plasma/ksystemstats/-/commit/fc8120f7500c3bfac1de23b6dc5d53e9eb34de06 Created attachment 186126 [details]
Sensors shown in 6.5.0 in Systemmonitor
Created attachment 186127 [details]
and shown in btop but 1min later
GPU0 in Picture of Systemmonitor is actually seen by nvidia-smi as GPU0 and has the Screens attached to it. It is using the GPU2 Sensors and now in 6.5.0 the VRAM used is displayed correctly, refering to btop. GPU1 in Picture is seen by nvidia-smi as GPU1 and has no screens attached to it. (Only for compute tasks here). It is using GPU1 Sensors in Systemmonitor. While Load, pwr and temp are shown correctly, the GPU1 Videomemory Used Sensor still is still showing erratic values. Nevermind. A Reboot and resetup of all sensors later it finally works as expected. *** Bug 500952 has been marked as a duplicate of this bug. *** |