powerstat options are as follow:
-a enable all statistics gathering options, equivalent to -c, -f, -g, -t and -H.
-b redo a sample measurement if a system is busy, the default for busy is considered less than 98%
CPU idle. The CPU idle threshold can be altered using the -i option.
-c gather CPU C-state activity and show the % time and count in each C-state at the end of the run.
-ddelay
specify delay in seconds before starting, default is 180 seconds when running on battery or 0
seconds when using RAPL. This gives the machine time to settle down and for the battery readings
to stabilize.
-D enable extra power stats showing all the power domain power readings. This currently only applies
to the -R RAPL option.
-f compute the geometric mean of all on-line CPU core frequencies. Unfortunately a CPU core is always
active to gather any form of stats because powerstat has to be running to do so, so these
statistics are skewed by this. It is best to use this option with a reasonably large delay (more
than 5 seconds) between samples to reduce the overhead of powerstat.
-g report GPU frequency of card0.
-h show help.
-H show histogram of power measurements.
-ithreshold
specify the idle threshold (in % CPU idle) to force a re-sample measurement if the CPU is less
idle than this level. This option implicitly enables the -b option.
-n no headings. Column headings are printed when they scroll off the terminal; this option disables
this and allows one to capture the output and parse the data without the need to filter out the
headings.
-p redo a sample measurement if any processes fork(), exec() or exit().
-r redo if system is not idle and any processes fork(), exec() or exit(), an alias for -p -b.
-R read power statistics from the RAPL (Running Average Power Limit) domains. This is supported by
recent Linux kernels and Sandybridge and later Intel processors. This only covers some of the
hardware in the machine, such as the processor package, DRAM controller, CPU core (power plane 0),
graphics uncore (power plane 1) and so forth, so the readings do not cover the entire machine.
Because the RAPL readings are accurate and available immediately, the start delay (-d option) is
defaulted to zero seconds.
-s this dumps a log of the process fork(), exec() and exit() activity on completion.
-S use standard averaging to calculate power consumption instead of using a 120 second rolling
average of capacity samples. This is only useful if the battery reports just capacity values and
is an alternative method of calculating the power consumption based on the start and current
battery capacity.
-t gather temperatures from all the available thermal zones on the device. If there are no thermal
zones available then nothing will be displayed.
-z forcibly ignore zero power rate readings from the battery. Use this to gather other statistics
(for example when using -c, -f, -t options) if powerstat cannot measure power (not discharging or
no RAPL interface).