radxa-docs · peterwang2050 · Dec 23, 2025 · Dec 22, 2025 · Dec 22, 2025 · Dec 22, 2025
@@ -135,13 +135,19 @@ gst-launch-1.0 v4l2src num-buffers=512 device=/dev/video11 io-mode=4 ! videoconv
 
 ## 引脚定义
 
-| Pin | Signal | Pin | Signal | Pin | Signal  | Pin | Signal  |
-| --- | ------ | --- | ------ | --- | ------- | --- | ------- |
-| 1   | DGND   | 9   | NC     | 17  | MCN     | 25  | SDA     |
-| 2   | MDN3   | 10  | DGND   | 18  | MCP     | 26  | NC      |
-| 3   | MDP3   | 11  | MDN1   | 19  | DGND    | 27  | RESET   |
-| 4   | DGND   | 12  | MDP1   | 20  | MCLK    | 28  | VCC3.3V |
-| 5   | MDN2   | 13  | DGND   | 21  | DGND    | 29  | VCC3.3V |
-| 6   | MDP2   | 14  | MDN0   | 22  | NC      | 30  | VCC5V   |
-| 7   | DGND   | 15  | MDP0   | 23  | PWDN/NC | 31  | VCC5V   |
-| 8   | NC     | 16  | DGND   | 24  | SCL     | 32  | ~       |
+| Pin | Signal | Pin | Signal | Pin | Signal    | Pin | Signal     |
+| --- | ------ | --- | ------ | --- | --------- | --- | ---------- |
+| 1   | DGND   | 9   | NC     | 17  | MCN       | 25  | SDA        |
+| 2   | MDN3   | 10  | DGND   | 18  | MCP       | 26  | NC         |
+| 3   | MDP3   | 11  | MDN1   | 19  | DGND      | 27  | RESET      |
+| 4   | DGND   | 12  | MDP1   | 20  | MCLK      | 28  | VCC3.3V    |
+| 5   | MDN2   | 13  | DGND   | 21  | DGND      | 29  | VCC3.3V    |
+| 6   | MDP2   | 14  | MDN0   | 22  | NC        | 30  | VCC5V / NC |
+| 7   | DGND   | 15  | MDP0   | 23  | PWDN / NC | 31  | VCC5V / NC |
+| 8   | NC     | 16  | DGND   | 24  | SCL       | 32  | ~          |
+
+:::tip 注意事项
+
+在瑞莎 13M 214 摄像头模组中，Pin 23 / 30 / 31 为悬空未连接脚位。
+
+:::
@@ -1,48 +1,67 @@
 ## NPU 简介
 
 瑞莎星睿 O6 / O6N 拥有高达 28.8 TOPS NPU (Neural Processing Unit) 算力，支持 INT4 / INT8 / INT16 / FP16 / BF16 和 TF32 类型的加速。
+
 此文档主要讲解用户如何使用 CIX P1 NPU SDK 运行基于 NPU 加速推理的人工智能模型和应用。包括模型的编译工具、工具链和一些常见模型案例的使用方法进行逐步讲解。
 
-## CIX 早鸟计划
+## 此芯 SDK
+
+我们需要访问[此芯开发者中心](https://developer.cixtech.com/)获取最新的此芯 AI 开发工具包（NeuralONE AI SDK）。
+
+此芯P1AI开发工具包，支持NPU等异构硬件加速，助力开发者开发端侧AI应用，实现高能效端侧AI推理。
+
+:::tip 此芯开发者中心
+
+此芯开发者中心包含软件 SDK、芯片手册、开发文档等资源。
 
-NPU SDK 包含 NOE Compiler 与 CIX AI Model Hub，要获取 SDK 需要通过申请 CIX 早鸟计划来获得。
-申请链接： [CIX 早鸟计划](https://earlybird.cixtech.com/#/)
+:::
+
+### 下载 SDK
 
-## 安装 NPU SDK (x86 Linux 环境)
+注册并登录[此芯开发者中心](https://developer.cixtech.com/)，在软件 SDK 内点击 NeuralONE AI SDK 选项的了解更多选项，会自动下载 SDK。
 
-通过 [CIX 早鸟计划](#cix-早鸟计划) 申请后，会有邮件发送到注册邮箱，请按邮件内容下载 SDK
+### 解压 SDK
 
 ```bash
-tar -xvf nor.tag.gz
+tar -xvf cix_noe_sdk_xxx_release.tar.gz
 ```
 
-:::tip
-CIX P1 NPU SDK 是安装在 X86 主机上，并要求 python 版本为 3.8。
+解压后的文件夹中包含以下内容：
 
-请用户准备好 python3.8 环境，推荐使用 [anaconda](https://www.anaconda.com/download)
-:::
+- cix-noe-umd_xxx_arm64.deb
+- cix-npu-driver_xxx_arm64.deb
+- CixBuilder_xxx-cp310-none-linux_x86_64.whl
+- env_setup.sh
+- npu_sdk_last_manifest_list.xml
+- requirements.txt
+
+### 安装 NPU 驱动
+
+进入解压后的文件夹，执行以下命令安装 NPU 驱动。
 
-### 安装 NOE Compiler
+```bash
+sudo dpkg -i ./cix-npu-driver_xxx_arm64.deb
+```
+
+### 安装 NOE 编译器
 
-NOE Compiler 用于模型的编译，将 ONNX 模型框架的模型格式转换为可以使用 NPU 进行加速推理的模型格式
+NOE 编译器用于模型的编译，将 ONNX 模型框架的模型格式转换为可以使用 NPU 进行加速推理的模型格式
 
 ```bash
-cd noe
 pip3 install -r requirements.txt
-pip3 install ./CixBuilder-6.1.2958.1-py3-none-any.whl
+pip3 install ./CixBuilder_xxx-cp310-none-linux_x86_64.whl
 ```
 
-## ~安装 NOE UMD (Orion O6 / O6N)~
+### 安装 NOE UMD
 
-:::tip
-~NOE UMD 包以 deb 安装包形式发布，如使用 python api，在创建虚拟环境时请继承系统依赖： **`python3 -m venv --system-site-packages .venv`**~
-:::
-~通过 [CIX 早鸟计划](#cix-早鸟计划) 申请后，会有邮件发送到注册邮箱，请按邮件内容下载 UMD (User Mode Driver) deb 包~
+```bash
+sudo dpkg -i ./cix-noe-umd_xxx_arm64.deb
+```
 
-~`bash
-sudo dpkg -i ./cix-noe-umd_0.01-1_arm64.deb
-~`~
+### 验证安装
 
-:::tip
-NOE UMD 已经通过预装的形式安装在 OS 里， 如使用 python api，在创建 python 虚拟环境时请继承系统依赖： **`python3 -m venv --system-site-packages .venv`** 以确保 libnoe 模块可被正确加载
-:::
+使用 cixbuild 命令验证 NOE 编译器是否安装成功。
+
+```bash
+cixbuild -v
+```
@@ -132,13 +132,19 @@ gst-launch-1.0 v4l2src num-buffers=512 device=/dev/video11 io-mode=4 ! videoconv
 
 ## Pin Definition
 
-| Pin | Signal | Pin | Signal | Pin | Signal  | Pin | Signal  |
-| --- | ------ | --- | ------ | --- | ------- | --- | ------- |
-| 1   | DGND   | 9   | NC     | 17  | MCN     | 25  | SDA     |
-| 2   | MDN3   | 10  | DGND   | 18  | MCP     | 26  | NC      |
-| 3   | MDP3   | 11  | MDN1   | 19  | DGND    | 27  | RESET   |
-| 4   | DGND   | 12  | MDP1   | 20  | MCLK    | 28  | VCC3.3V |
-| 5   | MDN2   | 13  | DGND   | 21  | DGND    | 29  | VCC3.3V |
-| 6   | MDP2   | 14  | MDN0   | 22  | NC      | 30  | VCC5V   |
-| 7   | DGND   | 15  | MDP0   | 23  | PWDN/NC | 31  | VCC5V   |
-| 8   | NC     | 16  | DGND   | 24  | SCL     | 32  | ~       |
+| Pin | Signal | Pin | Signal | Pin | Signal    | Pin | Signal     |
+| --- | ------ | --- | ------ | --- | --------- | --- | ---------- |
+| 1   | DGND   | 9   | NC     | 17  | MCN       | 25  | SDA        |
+| 2   | MDN3   | 10  | DGND   | 18  | MCP       | 26  | NC         |
+| 3   | MDP3   | 11  | MDN1   | 19  | DGND      | 27  | RESET      |
+| 4   | DGND   | 12  | MDP1   | 20  | MCLK      | 28  | VCC3.3V    |
+| 5   | MDN2   | 13  | DGND   | 21  | DGND      | 29  | VCC3.3V    |
+| 6   | MDP2   | 14  | MDN0   | 22  | NC        | 30  | VCC5V / NC |
+| 7   | DGND   | 15  | MDP0   | 23  | PWDN / NC | 31  | VCC5V / NC |
+| 8   | NC     | 16  | DGND   | 24  | SCL       | 32  | ~          |
+
+:::tip Note
+
+In the Radxa 13M 214 camera module, Pin 23 / 30 / 31 are not connected.
+
+:::
@@ -1,48 +1,67 @@
-## NPU Introduction
+## NPU Overview
 
-Radxa Orion O6 / O6N features up to 28.8 TOPS NPU (Neural Processing Unit) computing power, supporting acceleration for INT4 / INT8 / INT16 / FP16 / BF16, and TF32 data types.
-This document explains how to use the CIX P1 NPU SDK to run AI models and applications with NPU-accelerated inference. It includes step-by-step guides for model compilation tools, toolchains, and example implementations of common models.
+Radxa Orion O6 / O6N delivers up to 28.8 TOPS of NPU (Neural Processing Unit) performance and supports INT4 / INT8 / INT16 / FP16 / BF16, and TF32 acceleration.
 
-## CIX Early Bird Program
+This document explains how to use the CIX P1 NPU SDK to run AI models and applications that leverage NPU-accelerated inference. It covers the compilation tools, toolchains, and step-by-step guidance for common sample models.
 
-The NPU SDK includes NOE Compiler and CIX AI Model Hub. To obtain the SDK, you need to apply for the CIX Early Bird Program.
-Application link: [CIX Early Bird Program](https://earlybird.cixtech.com/#/)
+## CIX SDK
 
-## Install NPU SDK (x86 Linux Environment)
+Visit the [CIX Developer Center](https://developer.cixtech.com/) to download the latest CIX AI development toolkit (NeuralONE AI SDK).
 
-After applying through the [CIX Early Bird Program](#cix-early-bird-program), you will receive an email with download instructions for the SDK.
+The CIX P1 AI development toolkit supports heterogeneous hardware acceleration, including NPUs, enabling energy-efficient edge AI inference.
+
+:::tip CIX Developer Center
+
+The CIX Developer Center provides software SDKs, chip manuals, development guides, and more resources.
+
+:::
+
+### Download the SDK
+
+Register and sign in to the [CIX Developer Center](https://developer.cixtech.com/). Inside the Software SDK section, click **Learn more** under the NeuralONE AI SDK entry to start the download automatically.
+
+### Extract the SDK
 
 ```bash
-tar -xvf nor.tag.gz
+tar -xvf cix_noe_sdk_xxx_release.tar.gz
 ```
 
-:::tip
-The CIX P1 NPU SDK is installed on an X86 host system and requires Python 3.8.
+After extraction, the folder contains the following files:
 
-Please prepare a Python 3.8 environment. We recommend using [Anaconda](https://www.anaconda.com/download).
-:::
+- cix-noe-umd_xxx_arm64.deb
+- cix-npu-driver_xxx_arm64.deb
+- CixBuilder_xxx-cp310-none-linux_x86_64.whl
+- env_setup.sh
+- npu_sdk_last_manifest_list.xml
+- requirements.txt
+
+### Install the NPU Driver
+
+Change into the extracted folder and run the following command to install the NPU driver:
 
-### Install NOE Compiler
+```bash
+sudo dpkg -i ./cix-npu-driver_xxx_arm64.deb
+```
+
+### Install the NOE Compiler
 
-NOE Compiler is used for model compilation, converting models from ONNX framework format to a format that can be accelerated by the NPU for inference.
+The NOE Compiler converts ONNX models into a format optimized for NPU-accelerated inference.
 
 ```bash
-cd noe
 pip3 install -r requirements.txt
-pip3 install ./CixBuilder-6.1.2958.1-py3-none-any.whl
+pip3 install ./CixBuilder_xxx-cp310-none-linux_x86_64.whl
 ```
 
-## ~Install NOE UMD (Orion O6 / O6N)~
+### Install NOE UMD
 
-:::tip
-~The NOE UMD package is released as a .deb package. If using the Python API, please inherit system dependencies when creating a virtual environment: **`python3 -m venv --system-site-packages .venv`**~
-:::
-~After applying through the [CIX Early Bird Program](#cix-early-bird-program), you will receive an email with instructions to download the UMD (User Mode Driver) .deb package~
+```bash
+sudo dpkg -i ./cix-noe-umd_xxx_arm64.deb
+```
 
-~`bash
-sudo dpkg -i ./cix-noe-umd_0.01-1_arm64.deb
-~`~
+### Verify the Installation
 
-:::tip
-NOE UMD is pre-installed in the OS. If using the Python API, please inherit system dependencies when creating a Python virtual environment: **`python3 -m venv --system-site-packages .venv`** to ensure the libnoe module is loaded correctly.
-:::
+Use the `cixbuild` command to confirm that the NOE Compiler is installed correctly.
+
+```bash
+cixbuild -v
+```