Can it be said that the GPU power output from AGX Xavier is correct?

I am measuring the power of the xavier. Especially on the GPU. However, the measurement result is a sharp triangular waveform. In fact, we know that power consumption should be output in a square wave. In fact, the curious thing is that when power is consumed in a square wave shape, the maximum power cannot be determined because the power measurement may drop due to the end of the process even though the measured power consumption is not peaked. Is there a way to clearly see the pattern of energy consumption?

Where did you probe the “power consumption” on? Is it voltage or current? The command line “tegrastats” can show the power value of each domain.

it’s current. I measured xavier’s gpu current at /sys/bus/i2c/devices/1-0040/iio_device/in_power0_input. Measurement is very fast, about 2.3ms, and a sharp triangular waveform appears when the measured waveform is plotted.

I mean which point on board are you probing? Do you have the screenshot of that and waveform?

I am measuring the gpu current using the path mentioned above, and this is a picture of the measurement result, and an enlarged picture of the rising front part is also attached.


The measurement I want is not a blue dotted line, but a purple line. This graph was taken from a paper entitled’Measuring GPU Power with the K20 Built-in Sensor’.
cc

So you are reading the value from sw node not with a hardware equipment? It is the output data from a onboard power monitor with fixed sample rate and average value. No problem with that, and the command line “tegrastats” can show the power value too.

Is it the result of the sample rate of the internal software node?
Then, we can’t get a square wave with the SW node… Okay. Thank you!

Yes, it is. To get real power consumption waveform, you need to use an oscilloscope to measure the voltage and voltage drop of the resistor in serial on the VDD_GPU line on module board.