I was following the Nvidia/TensorRT github page to convert my onnx model into tensorrt engine.
While during the process, jetson started to throw notifications
“System throttled due to overcurrent”.
While I did search online, I could not find any solution for Jetpack > 5,0.
Can you please help me with this?
Also please let me know if sudo nvpmodel -m 0 is really the best choice for jetson xavier nx when I want maximum performance?
We want to reproduce this issue internally.
Could you share a sample to reproduce the behavior?
Ok, here it goes.
First pull and run docker image.
sudo docker pull nvcr.io/nvidia/l4t-pytorch:r35.2.1-pth2.0-py3
sudo docker run -it --rm --runtime nvidia --network host -v /home/user/project:/location/in/container nvcr.io/nvidia/l4t-pytorch:r35.2.1-pth2.0-py3
sudo nvpmodel -m 2
pip install -r requirements
Since I had already created a converted.onnx in my ubuntu’s docker, I directly did
python3 build_engine.py --onnx converted.onnx --engine engine.trt --precision fp16
If something is not clear. I apologize.
Please let me know.
We will check it and update.
I just want to confirm one thing.
I saw this solution for jetpack 4.5, will this work for jetpack 5.1?
# echo 5000 > /sys/devices/c250000.i2c/i2c-7/7-0040/iio:device0/crit_current_limit_0
Although I did check that c250000 folder is not present in devices.
The command rise the OC limit to 5A.
But the default setting in JetPack 5.1 is already 5A.
We are checking this issue internally.
Will share more information with you later.
Could you share the
converted.onnx with us as well?
It is quite alright since I was able to create my tensorrt engine anyways.
Also I think that since my adapters output current limit is 2.37 A, maybe that why it was showing the message?
Anyways, Thank You!
Hello, I am also having this issue on my Jetson Xavier NX. However, I constantly get error messages about system throttle on any power mode I try, without running any resource intensive processes.
We had some internal discussions about this issue.
With some GPU-intensive jobs, the XavierNX might reach the overcurrent limit.
The maximum OC value on the XavierNX devkit is 5A. So you will see the warning message.
We also have some other platforms that can run over 5A such as Xavier, and Orin.
Please make sure you have upgraded to JetPack 5.1.
@AastaLLL Thank You for this information.
How did you end up solving the overcurrent problem when running TensorRT?
Unfortunately, it shows even now, but it is because of the gpu-intensive job. (working with tensorrt and ros)
this is a real problem… i upgraded from tx2 to xavier nx specifically for the fps but it keeps throttled what are possible solutions
im running jetpack 5.1
I’ve found the file that is related to oc limit feature in ‘/sys/devices/platform/c250000.i2c/i2c-7/7-0040/hwmon/hwmon5’ folder
curr1_crit curr2_crit curr3_crit curr4_crit in1_input in3_enable in6_input power subsystem
curr1_crit_alarm curr2_crit_alarm curr3_crit_alarm curr4_crit_alarm in1_label in3_input in7_input samples uevent
curr1_input curr2_input curr3_input curr4_input in2_enable in3_label in7_label shunt1_resistor update_interval
curr1_max curr2_max curr3_max device in2_input in4_input name shunt2_resistor
curr1_max_alarm curr2_max_alarm curr3_max_alarm in1_enable in2_label in5_input of_node shunt3_resistor
when I do # cat curr1_crit , it returns 5000 I think this indicates 5A
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.