You are not logged in.
I'm using a laptop with a Nvidia 3060 GPU. I discovered that when I run
nvidia-smi
to get the temperature of my GPU, the GPU will be "waken", resulting in a significant amount of power consumption. I've also tried the same command in Windows and got the same outcome. However, I noticed that the "Task manager" in Windows is able to tell the GPU temperature without "waking" it. So, I'm wondering if there is a similar way in linux to do the same?
Last edited by elight (2024-11-18 08:14:50)
Offline
I noticed that the "Task manager" in Windows is able to tell the GPU temperature without "waking" it.
How?
I'd check /sys/bus/pci/devices/0000:??:??.0/power/runtime_status and if the GPU is active, seize the opportunity to draw the temperature.
If it falls inactive, check the temperature one last time a couple of seconds (30-60) later and assume that value to be constant until the GPU activates again.
Maybe poll the temp of the inactive GPU at a really low frequency (every 5 minutes or so)
Iff windows ends up doing sth. like that you would probably have not noticed?
Offline
I noticed that the "Task manager" in Windows is able to tell the GPU temperature without "waking" it.
How?
I'd check /sys/bus/pci/devices/0000:??:??.0/power/runtime_status and if the GPU is active, seize the opportunity to draw the temperature.
If it falls inactive, check the temperature one last time a couple of seconds (30-60) later and assume that value to be constant until the GPU activates again.
Maybe poll the temp of the inactive GPU at a really low frequency (every 5 minutes or so)Iff windows ends up doing sth. like that you would probably have not noticed?
You are right, I discovered that Windows only updates its temperature value when nvidia-smi activates the GPU. Thanks for help.
Offline
Please always remember to mark resolved threads by editing your initial posts subject - so others will know that there's no task left, but maybe a solution to find.
Thanks.
Offline