I’d like to know if it is possible to measure GPU temperature in Linux. (Reading GPU core temperature reading)
I’ve seen previous posts about running “nvidia-settings” and look at the GPU temperature.
Is there any way to check the temperature when x-windows is started (setting /etc/inittab runlevel 3)
Is there any way to dump out the temperature readings to a separate text file ?
The nvidia-settings control panel shows the gpu core temperature, so does the third-part nvclock utility. There isn’t a sensor logging package that I know of with gpu capabilities, but nvclock and about 10 lines of bash script should take care of it.
Yes and no. nvclock is a command line utility, so you can run it from any shell, but it seems that whatever interface it uses only loads when the X11 modules are loaded and X11 is running (or at least that is the case with the hardware I have access to).
Right after the initial installation of the 190.18 driver for nvidia, I could see the thermal monitor readings from the nvidia-settings.
But after a reboot, now I can’t see the thermal monitor readings from the nvidia-settings panel. When I run nvidia-smi -lsa to probe the GPU core temperature, I get the following message. Rebooting or re-installing the cuda driver didn’t work.
Does anyone have any idea on how to troubleshoot this problem?
$ nvidia-smi -lsa
==============NVSMI LOG==============
Timestamp : Tue Oct 6 13:42:26 2009
GPU 0:
Product Name : GeForce GTX 285
PCI ID : 5e310de
Failed to read GPU temperature!
Temperature : 0 C
GPU 1:
Product Name : GeForce GTX 285
PCI ID : 5e310de
Failed to read GPU temperature!
Temperature : 0 C
GPU 2:
Product Name : GeForce GTX 285
PCI ID : 5e310de
Failed to read GPU temperature!
Temperature : 0 C