sar is a great tool to log various resource usage metrics, regarding CPU, memory, I/O, network...
...but not GPU?
I won't be remotely original by stressing the importance of logging GPU usage metrics (load, temperature, etc.) for some of us. Is there a way to have sar aware of those probes and log their values along the other metrics?
No. Sysstat is not aware of GPUs, so it can't log anything about them.
You could change that yourself – add a GPU-specific to sysstat's source code, or make nvidia write that code or pay someone else to write it. Especially, write regular emails to your GPU sales representative complaining that nvidia's kernel interface doesn't support the usual sysfs / procfs interfaces used by other resources in the Linux kernel.
