JETSON ORIN NX 16GB MODULE Power problem

Sorry to bother you.
I’m Asus IoT EE Leader, we currently have some issues regarding the Orin NX 16G SOM that we need your help to clarify.

  1. Regarding the issue of sufficient wattage for 40W power.
    We are using the Jetson Power GUI tool to monitor power, but after aging, both the CPU and GPU loading reach 100%, while the “VDD IN” is only 33W.
    Is this correct? Is the SOM limiting the power to prevent the system from reaching the full 40W?
    We try to modify the SOM’s instantaneous and average current limit, that power can up to 39W. May i ask if there are any concerns with modifying the current limit and has any impact items?

  2. Additionally, do you have any recommended methods for the aging tool?

Hi kd_tsai,

Are you verifying 40W power with the devkit or custom board for Orin NX 16G?
What’s the Jetpack version in use?

It may depend on your test tool.
How did you perform the test?

Do you modify the sysfs node manually?
If so, we don’t suggest doing that to prevent the damage to the SoM.
Please also take care of the cooling method to prevent high temperature when you are verifying with high power.

We’ve tried to run matrixMulCUBLAS from cuda-samples to perform stress test on the devkit and it could achieve 37W.

Dear,

thank you for your reply.

  1. We’ve tried to run matrixMulCUBLAS (–sizemult=8) to perform stress test, the system only achieve 35W.
    Our NV Kernel version is 36.4.3. May I ask that version you are testing?
  2. Yes, we modify the sysfs nodes manually. So NV is don’t suggest we to modify it right?

Yes, we were also verifying Jetpack6.2(L4T r36.4.3) with super mode configuration.
Have you referred to Jetson/L4T/TRT Customized Example - eLinux.org and apply the patches there before the test?

Correct, the default limit is verified internally and we don’t suggest customer modifying them.

It may be caused from OC throttled is triggered.
You can simply run the following command.

# grep "" /sys/class/hwmon/hwmon*/oc*

Please check before the workload is started and after the workload is completed

Dear Kevin

Thank you for your helping.

  1. The intercepted log “Burnin_log” I provided here should be that the stress testing tool pulled up all CPU and GPU at once and caused the OC protection to be locked, which is why it is only about 33W.

Burnin_log.txt (240 Bytes)

  1. Sorry, is it not recommended to manually adjust the OCP limit for the NV image by default, right?
  2. Could you help to provide the any dimensional information of SOM entity?

Dear

Sorry to bother you.
May I ask if the VDD_IN on the SOM has OVP to protect module?

Correct, we do not suggest modifying the limit of INA3221 (especially increase the threshold to allow more power).

Please refer to Jetson Orin Nano Series, Jetson Orin NX Series and Jetson AGX Orin Series — NVIDIA Jetson Linux Developer Guide for details about Platform Power and Performance.

OC1 indicates the under voltage.
OC2 indicates average power
OC3 indicates instantaneous power
You can find more details in above document I shared.

From the result you shared, it seems you hit OC3 issue.