Monitoring the performance of a GPU is not pretty easy and it depends on the GPU. In this case we’re monitoring a NVIDIA Titan X on a Windows Machine, mainly for statistics. However there are mostly two ways to get this done.
- using NVIDIA smi or
- using OpenHardwareMonitor
OpenHardwareMonitor is a small tool to monitor various hardware sensors like fans, temperature, load and stuff like this. The really special about this tool is: It outputs the values via WMI.
First we would have to be sure that WMI and OpendHardwareMonitor is up and running on our target. Then we would see what the WMI Object actually returns:
It will return a lot of stuff, but we’re looking for the GPU load or other sensors we would like to monitor. Now we can send a specific query to get the value:
As we would like to use this as Custom Sensor in PRTG, we well have to provide the credentials in the script. In my case the monitored computer provides 4 GPUs in total and we want to monitor all of them. To make it easier to handle I decided to use only one sensor with 4 channels in PRTG using the custom XML sensor, which needs to return the values formated in a special scheme:
Next we copy the ps script to
C:\Program Files (x86)\PRTG Network Monitor\Custom Sensors\EXEXML on the probe.
Finally we have to configure the Sensor: As parameters we will use the placeholders provided by PRTG:
'%host' '%windowsuser' '%windowspassword' '%windowsdomain'