Power Consumption measurement during training

I would like to measure the CPU/GPU (POM_5V_CPU/POM_5V_GPU) power consumption while training a machine learning model on the Jetson Nano (P3450). I am using the tegrastats utility with this command:

sudo tegrastats --interval 1000 > tegra_dump.log

Two questions:

  1. When using the --interval 1000 flag, does this mean that the second value reported is the average current over the 1000 ms?
  2. Is there a better/recommended way to measure power consumption

Related forum posts:

Hi ali.jahangiri,

Are you using the devkit or custom board for Jetson Nano?

Correct, it is indeed the average readings over that 1000ms interval.

Maybe you can also try using jtop tool.

$ sudo -H pip install -U jetson-stats

Or read from sysfs node as following:

$ cat /sys/bus/i2c/drivers/ina3221x/6-0040/iio:device0/in_power0_input

Are you using the devkit or custom board for Jetson Nano?

Yes, I am using the original Devkit

Or read from sysfs node as following:
$ cat /sys/bus/i2c/drivers/ina3221x/6-0040/iio:device0/in_power0_input

Would the Power Monitor Information be the part of the NVIDIA Jetson Linux Developer Guide I should refer to on how to access the INA3221 readings?

Thanks Kevin!

Yes, you can refer to that part for those readings.