Jetson Orin Nano device hangs a lot

I’m using a Jetson Orion Nano device for two web apps. One employs the YOLOv5 model, while the other utilizes the TRT_pose model. I only run one application at a time. However, when I start either of the applications, the model loading process takes a considerable amount of time. Interestingly, I’ve also tested the same app on another Jetson Nano device, where it runs flawlessly. It seems that the device I’m currently using is significantly slower. It responds sluggishly, and eventually, the system freezes, even when no applications are running.

CPU: Athlon
RAM: 8GB

Please help

Hi,

Do you build your application with Orin Nano architecture (sm_87)?

A possible reason is that the PTX is generated with other architecture (ex. 53 for Jetson Nano).
So every time you try to launch the application, it triggers JIT compiling and takes time.

Thanks.

HI. Sorry. I didn’t get you. I am new to Nvidia. All I can say is that the application is in node.js. And I did not train the yolo model using the jetson device. I trained it on another machine. But it is the same yolo model that I am using on other 2 jetson device that I have and those two devices are working perfectly fine and the model loads really quick on those. Overall the entire device is very very slow. It responds very slow even when I try to open a the file manage

Hi,

How do you infer the model?
If you run it with TensorRT, please create the engine file on the Orin Nano directly.
If there is any CUDA implementation, please also check if you have compiled it with the sm_87 architecture to prevent the JIT issue.

Thanks.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.