I have been testing my company’s Firmware Application that runs with YOLOv4 Model for live view object detection on Jetson AGX Orin devices with the following specs:
Jetpack 5.0.2
Python 3.8.10
CUDA 11.4.315
TensorRt: 8.6.3
OpenCV 4.4.0 with Cuda
Vulkan: 1.3.203
The following Drive Link will consist of live video capture of GPU performance on Jtop while running the firmware: ANS record - Google Drive
It seems that the GPU performance is not stable and at some point even stops running and goes idle.
Has anyone encountered performance issues like me and is there anyway I can boost the core clock of Jetson Orin?