本文へスキップ
SDK Version: Next

Building DX-Stream on Orange Pi 5 Plus

This guide provides comprehensive instructions for building and running DX-Stream on the Orange Pi 5 Plus single-board computer with DEEPX DX-M1 NPU for AI accelerator.

The Orange Pi 5 Plus is a powerful ARM-based development board designed for high-performance edge computing. When paired with the DEEPX DX-M1 NPU, it provides significant acceleration for AI and computer vision applications. This guide covers the complete setup process from SD card preparation to running DX-Stream applications.

DX-Stream Building Process

The overall construction and validation process is broken down into four sequential steps.

  • Step 1: SD Card Setup
  • Step 2: DX-RT NPU Driver Installation
  • Step 3: Build DX-Stream
  • Step 4: Running Demo Applications

Prerequisites​

This section outlines the essential hardware and software components required for building and running DX-Stream on the Orange Pi 5 Plus with the DEEPX DX-M1 NPU. Ensuring these requirements is crucial for a successful installation.

Hardware Requirements​

  • Orange Pi 5 Plus development board
  • DEEPX DX-M1 NPU module (M.2 form factor)
  • MicroSD card (32GB or larger, Class 10 recommended)
  • Power supply (5V/4A minimum)
  • Host Linux machine for SD card preparation

Software Requirements​

  • 7-Zip or similar extraction utility
  • SD card flashing tool or dd command access

[Step 1] SD Card Setup​

Prepare the base Ubuntu OS environment.

[Step 1.1] Download & Extract Ubuntu Image​

Download the official Orange Pi 5 Plus Ubuntu 22.04 image:

Extract the downloaded 7z archive:

7z x Orangepi5plus_1.2.0_ubuntu_jammy_desktop_xfce_linux6.1.43.7z
NOTE

If 7-Zip is not installed, install it using:

# Ubuntu/Debian
sudo apt install p7zip-full

# CentOS/RHEL
sudo yum install p7zip

[Step 1.2] Flashing the SD Card​

WARNING

Replace /dev/sda with your actual SD card device. Use lsblk or fdisk -l to identify the correct device.

sudo umount /dev/sda

sudo dd if=./Orangepi5plus_1.2.0_ubuntu_jammy_desktop_xfce_linux6.1.43.img of=/dev/sda bs=4M status=progress

sync

Alternative Methods

  • Raspberry Pi Imager: Cross-platform GUI tool
  • Balena Etcher: User-friendly flashing utility
  • Win32DiskImager: Windows-specific tool

[Step 1.3] Initial Boot & Credentials​

  • (1) Insert the SD card into the Orange Pi 5 Plus
  • (2) Connect power supply and peripherals
  • (3) Power on the device

Default Credentials (if applicable)

  • Username: orangepi
  • Password: orangepi

[Step 2] DX-RT NPU Driver & Runtime Installation​

Install the necessary drivers and runtime for the DEEPX NPU accelerator.

All following steps should be performed on the Orange Pi 5 Plus itself after successful boot.

[Step 2.1] Updating Linux Kernel Headers​

Install the required kernel headers for driver compilation:

cd /opt
sudo apt install ./linux-headers-current-rockchip-rk3588_1.2.0_arm64.deb
NOTE

The kernel headers package should be pre-installed in the Orange Pi image. If missing, contact DEEPX support.

(Optional) Configure Git

If you encounter SSL certificate issues with GitHub:

git config --global http.sslVerify false
NOTE

This disables SSL verification globally. For production use, consider configuring proper certificates instead.

[Step 2.2] Compiling NPU Linux Driver​

cd ~
mkdir -p dxnn && cd dxnn
git clone https://github.com/DEEPX-AI/dx_rt_npu_linux_driver.git
cd dx_rt_npu_linux_driver/modules
./build.sh
sudo ./build.sh -c install

What this does:

  • Downloads the NPU kernel driver source code

  • Compiles the driver for the current kernel

  • Installs the driver modules and creates device nodes

[Step 2.3] Installing DX-RT Runtime​

cd ~/dxnn
git clone https://github.com/DEEPX-AI/dx_rt.git
cd dx_rt
./install.sh --all
./build.sh

Build Options:

  • --all: Installs all dependencies and development tools

  • Alternative: ./install.sh --minimal for runtime-only installation

[Step 2.4] Verifying Installation (dxcli)​

Test the NPU driver and runtime installation:

dxcli -s

Expected Output:

DXRT v3.0.0
=======================================================
 * Device 0: M1, Accelerator type
---------------------   Version   ---------------------
 * RT Driver version   : v1.7.1
 * PCIe Driver version : v1.4.1
-------------------------------------------------------
 * FW version          : v2.1.4
--------------------- Device Info ---------------------
 * Memory : LPDDR5 6000 MHz, 3.92GiB
 * Board  : M.2, Rev 1.5
 * Chip Offset : 0
 * PCIe   : Gen3 X4 [01:00:00]

NPU 0: voltage 750 mV, clock 1000 MHz, temperature 40'C
NPU 1: voltage 750 mV, clock 1000 MHz, temperature 40'C
NPU 2: voltage 750 mV, clock 1000 MHz, temperature 40'C
dvfs Disabled
=======================================================

Quick Diagnostics (If Device is Not Detected)

If the dxcli -s command fails to detect the hardware, execute the following diagnostic checks sequentially.

# 1. Verify if the DEEPX NPU kernel driver module is loaded into the system
lsmod | grep dx

# 2. Inspect kernel system logs for any DEEPX driver initialization or PCIe handshake errors
dmesg | grep -i deepx

[Step 3] Building DX-Stream Framework​

Compile and install the DX-Stream GStreamer framework and plugins.

[Step 3.1] Installing Build Dependencies​

The Orange Pi 5 Plus Ubuntu image includes pre-configured GStreamer packages with Rockchip hardware acceleration support. Install the additional required packages:

sudo apt-get update
sudo apt install -y python3-pip ninja-build
sudo pip install meson
sudo apt-get install -y libeigen3-dev libjson-glib-dev librdkafka-dev libmosquitto-dev libyuv-dev

Package Descriptions

  • python3-pip, ninja-build, meson: Build system components
  • libeigen3-dev: Linear algebra library for computer vision
  • libjson-glib-dev: JSON parsing library for GLib
  • librdkafka-dev: Apache Kafka client library
  • libmosquitto-dev: MQTT broker client library
  • libyuv-dev: YUV color space conversion library

[Step 3.2] Cloning & Compiling DX-Stream​

cd ~/dxnn
git clone https://github.com/DEEPX-AI/dx_stream.git
cd dx_stream
./build.sh

Build Process:

  • Configures Meson build system
  • Compiles all GStreamer plugins
  • Builds sample applications and utilities
  • Installs plugins to system directories

Build Time: Approximately 10-15 minutes on Orange Pi 5 Plus

[Step 3.3] Verifying GStreamer Plugins​

Check if DX-Stream plugins are properly installed:

gst-inspect-1.0 | grep dx

Expected Output:

dxstream:  dxgather: DX Gather
dxstream:  dxinfer: DX Inference
dxstream:  dxmsgbroker: DX Message Broker
dxstream:  dxmsgconv: DX Message Converter
dxstream:  dxosd: DX On-Screen Display
dxstream:  dxpostprocess: DX Post-process
dxstream:  dxpreprocess: DX Pre-process
...

[Step 4] Running Demo Applications​

Verify the complete system functionality and NPU acceleration.

[Step 4.1] Setting up Model & Video Assets​

cd ~/dxnn/dx_stream
./setup.sh

[Step 4.2] Executing Demo Pipeline​

./run_demo.sh

Available Demo Options:

  • Object detection with YOLO models
  • Face detection and recognition
  • Pose estimation
  • Real-time video processing

For detailed demo instructions, refer to the Installation Guide.

NOTE

Hardware Acceleration. The Orange Pi 5 Plus image includes optimized drivers for:

  • Video Decode/Encode (VPU): Rockchip RK3588 VPU
  • GPU Acceleration: Mali-G610 MP4
  • NPU Acceleration (AI): DEEPX DX-M1 (25 TOPS)

For maximum performance, ensure your pipelines utilize these specific hardware accelerators.

Troubleshooting & Diagnostics​

This section provides guidance for resolving common issues encountered during the DX-Stream installation or runtime.

Hardware & Plugin Issues (NPU, GStreamer Cache)​

NPU Not Detected

Verify the NPU module is correctly recognized and the driver is loaded.

# Check PCIe connection
lspci | grep -i deepx

# Verify driver loading
dmesg | grep -i dx_rt

GStreamer Plugin Not Found

If DX-Stream plugins are not registered, manually refresh the plugin cache.

# Refresh plugin cache
gst-inspect-1.0 --plugin-path=/usr/local/lib/gstreamer-1.0

Build & Performance Issues​

Build Failures

If the build process fails, clean the environment and rebuild.

# Clean and rebuild
./build.sh

Performance Issues

Monitor hardware status if performance is slower than expected.

  • Verify NPU Status
dxcli -s
  • Monitor CPU usage
htop
NOTE

Next Steps. fter successful installation:

Last updated: