The measurement was done by using the power tool GUI
4. The biggest power measurement was
VDD_IN-IN_POWER = 20224
VDD_CPU_GPU_CV-IN_POWER = 13454
VDD_SOC-IN_POWER = 2944
5. According to Xavier NX spec the maximum power is 20 W and according to document “NVIDIA Jetson Xavier NX Thermal Design Guide” TDG-09774-001_v1.4 the total power is 15.80 W.
the document can be found here: Jetson Download Center | NVIDIA Developer
I have the following questions:
From jtop I understand that for whole jetson’s power consumption I need to summarize all 3 power values above. So I actually got 20224 + 13454 + 2944 = 36.622 W.
What is the right way to calculate the Jetson’s power consumption from results of the power tool?
VDD_IN is the source of all power rails of module.
Real power consumption depends on use case and nvpmodel. Once nvpmodel is fixed, the total power will be limited to not exceed the threshold.
Thanks @Trumany,
I’m a teamate of @userjetson
Your answer is very clear but it is little bit confusing us based on our past experience which started since TX2 series.
We want to be sure that we are don’t miss anything, so, if it is OK I will want to desribe our understanding in more details.
Please help us to be sure that we perfectly understand how to measure in general any Jetson (Orin, Xavier, etc.) power consumption, specifically the Xavier NX.
Based on the document [Log in | NVIDIA Developer](https://Jetson TX2 Series Thermal Design Guide)
Paragraph 5.3, the insrtuctions for SW to measure the overall Jetson power consumtion is to sum all rails values - See the example script which is part of the paragraph:
for ((i = 0; i < ${#RAILS[@]}; i++)); do
pwr_sum[$i]=$((${pwr_sum[$i]} + $(cat ${node[$i]}))) &&
pwr_count[$i]=$((${pwr_count[$i]} + 1))
done
*Please pay attention that the VDD_IN rail is part of the TX2 rails list but stil the script sum all rails.
VDD_IN is detailed as the VDD_GPU in Table 5-1.
We compared this techniqe to the Python Jtop utility and verify that it reports each rail seperatly but its final result is also the sum of all rails.
Since then, for each Jetson we used to sum all rails values in our benchmarks.
So, the questions are:
Is it true that NVIDIA measure and report only the VDD_GPU in their benchmarks reports?
For formal benchmarks we are doing using the NVIDIA evaluations kit before we move to our production module which integrate the jetson on it with a specific carrier board, in order to have the most accuracy evaluated power consumption level, is it right to measure only the VDD_GPU or shall it includes all rails sum as an evaluated power measurement prediction for our dedicated overall module which will host the Jetson but includes also a carrier board?
As you can see in figure 5-1 in TDG, VDD_IN is the sum of all other rails. For the code, as said: This is an example script that can report the average power consumed by each of the 6 rails.
Thanks for the clarification with the figure (unfortunately this figure isn’t part of Xavier NX document).
Our remaining question is what rail do we need to take in our consideration, for understanding how much power same application will take on Jetson model instead of measured Jetson developer kit?
From what we understood, the VDD_IN contains power of all the developer kit components, including these which are not part of the Jetson module.
For better understanding, the VDD_IN of module (you can get that based on Xavier NX carrier board schematic) is the sum of all power supplies of module.