AMDGPU Monitor for KSysGuard

Python 3 script that allows user to monitor AMD GPUs in KSysGuard (AMDGPU driver only)

Features

• calculates average values of every sensor (looking at you gpu_busy_percent)
• supports multiple GPUs
• detects min and max values of sensors automatically
• doesn't require any special dependencies
• no root needed
• respects Update interval set in Tab properties in KSysGuard

NOTE: Some sensors (notably GPU & VRAM Voltages) require amdgpu.ppfeaturemask=0xfffd7fff boot parameter - more info

Requirements

• AMD GPU using AMDGPU driver
• Python 3 (tested with 3.8)
• read permission for files in /sys/bus/pci/drivers/amdgpu/<PCI_SLOT>/*
• having the most recent version of the Linux Kernel is highly recommended (especially for newer GPUs)

Available sensors

Based on Linux Kernel Documentation:

Sensor	Unit
GPU Usage	%
GPU Clock	MHz
GPU Voltage	mV
VRAM Usage	MiB
VRAM Clock	MHz
VRAM Voltage	mV
GTT Usage	MiB
Temperature	°C
Power	W
Fan	RPM
Fan PWM
SoC Clock (>= Vega10)	MHz
DCEF Clock (>= Vega10)	MHz
F Clock (>= Vega20)	MHz

Available arguments

Argument	Type	Description
--tick	float	Time (in sec) how often should the sensor data be collected
--logging	bool	Enables logging - use it when something isn't working. Do not use this argument when adding the script to KSysGuard!

Usage

1. Make sure the sensor.py file has an executable permission
2. Open KSysGuard
3. Open File -> Monitor Remote Machine... and fill the window appropriately:
   1. Host: AMDGPU (or whatever you want)
   2. Connection type: Custom command
   3. Command: <PATH TO sensor.py> [arguments]

If everything went OK, new sensors should be available in the Sensor Browser. Then just grab the sensor from the list and drop it in a desired place.

HINT: If the Monitor Remote Machine... is missing, create new tab.

Issues

• If you have a GPU from Vega 20 series or later, VRAM Voltage sensor might be missing - with these GPUs the driver reports some data (pp_od_clk_voltage file to be exact) in a different way and I haven't figured out (yet) a way how to detect it. Check Issue if you would like to help!
• If a sensor is missing or its values are -1 or 0, it means that the script failed to read and/or parse a file - please read Testing and include the output (with errors from --logging argument) when creating an Issue. Also, please include Linux Kernel version you're using.

Testing

It's possible to test the script manually by executing it in the terminal. If you do it with a --logging argument, you might see some errors get printed out after running it. Include them when creating an Issue!

Either way, the script now should print out:

ksysguardd 1.2.0
ksysguardd>

1. Type monitors to get list of all available monitors (in PCI_SLOT/SENSOR format)

ksysguardd> monitors
0000:01:00.0/gpu_usage  integer
0000:01:00.0/gpu_clock  integer
[...]

2. Now type PCI_SLOT/SENSOR? to get info about a specific sensor - it should print out monitor info in a name min max unit format. This command is required for the one from next step to work!

ksysguardd> 0000:01:00.0/gpu_usage?
GPU Usage       0       100     %

3. Now type PCI_SLOT/SENSOR to get current value of the sensor

ksysguardd> 0000:01:00.0/gpu_usage
8

TODO

• Add support for Vega 20 series VRAM voltage (Issue)
• Add support for Northbridge voltages (Issue)
• Add support for non-average values
• Add support for multiple temperature and fan sensors
• Add support for multiple hwmons (I'm not sure if it's even needed)
• Improve performance and code quality (any pointers are welcome)
• Add support for more units (if requested)