Deployment on Orin AGX 64G

[edit_or]

Please provide complete information as applicable to your setup.

**• Hardware Platform (Jetson / GPU)**Orin AGX 64G
• DeepStream Version7.0
I have a few queries since I have a project using Oring AGX 64G.
(1)How do I know that how many Resnet50 object detection type of models can run concurrently in Orin AGX 64G. How can I check?
(2)User can select different types of detection to run, like people counting, vehicle detection, etc. I am thinking that we’ll have front-end GUI to select different types of detection application to run, then respective inference model have to be uploaded and run. Which architecture will be a good to approach like triton server or nvinfer?
(3)Display will be in QT window. I can’t display using OSD because GUI has dedicated display window for each application. How can I send images to QT window from deepstream-app? Say, four detection models are running concurrently, how all these four pipelines update images to QT window to a specific display slot?

[Fiona.Chen]

@edit_or

(1)How do I know that how many Resnet50 object detection type of models can run concurrently in Orin AGX 64G. How can I check?

If only consider the models, please use “trtexec” tool to measure the performance.
Developer Guide :: NVIDIA Deep Learning TensorRT Documentation
TensorRT/samples/trtexec at release/10.2 · NVIDIA/TensorRT (github.com)

@edit_or

(2)User can select different types of detection to run, like people counting, vehicle detection, etc. I am thinking that we’ll have front-end GUI to select different types of detection application to run, then respective inference model have to be uploaded and run. Which architecture will be a good to approach like triton server or nvinfer?

Both are fine since nvinfer and nvinferserver are all accelerated by Nvidia GPU.

DeepStream is just a SDK, the GUI design depends on your own application.

@edit_or

(3)Display will be in QT window. I can’t display using OSD because GUI has dedicated display window for each application. How can I send images to QT window from deepstream-app? Say, four detection models are running concurrently, how all these four pipelines update images to QT window to a specific display slot?

You may consider nvdrmvideosink for display. Accelerated GStreamer — NVIDIA Jetson Linux Developer Guide 1 documentation
Please consult QT for how to integrate the window with DRM.

[edit_or]

Yes thanks @Fiona.Chen
(1)For concurrent running of models on Nvidia devices, which parameters are important to know that how many model the device can accept for running concurrently. Is it GPU memory usage for each model or sth more?

(3)Is there a sample c/cpp deepstream sample using DRM sink?

[Fiona.Chen]

@edit_or

(1)For concurrent running of models on Nvidia devices, which parameters are important to know that how many model the device can accept for running concurrently. Is it GPU memory usage for each model or sth more?

The GPU compute time and GPU memory usage.

@edit_or

(3)Is there a sample c/cpp deepstream sample using DRM sink?

Please refer to Accelerated GStreamer — NVIDIA Jetson Linux Developer Guide 1 documentation and DeepStream SDK FAQ - Intelligent Video Analytics / DeepStream SDK - NVIDIA Developer Forums

[yingliu]

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks