SmartFly info PCIe Gen3 AI Accelerator Card (CRL-G116U-P3DF) User Manual

1. Introduction

This manual provides essential information for the installation, operation, and maintenance of your SmartFly info PCIe Gen3 AI Accelerator Card, model CRL-G116U-P3DF. This card is designed for Edge AI Inference, leveraging Google Coral Edge TPU M.2 modules to deliver powerful AI inference capabilities. It is compatible with standard PCI Express Gen 3 x16 slots and features an optimized thermal design for stable performance under high loads.

Figure 1: Front view of the AI Accelerator Card, showcasing its dual cooling fans and PCIe interface.

Figure 2: Product overview detailing the card's features, including support for Google Edge TPU modules and its robust thermal solution.

2. Safety Information

Please read and follow these safety guidelines to prevent damage to the device or injury to yourself:

• Always disconnect power from your computer before installing or removing any internal components.
• Handle the card by its edges to avoid touching sensitive components.
• Ensure proper grounding to prevent electrostatic discharge (ESD). Use an anti-static wrist strap if available.
• Install the card in a well-ventilated computer case to prevent overheating.
• Do not expose the card to moisture or extreme temperatures.
• Keep out of reach of children.

3. Package Contents

Verify that all components are present in your package:

• 1 x AI Accelerator PCIe Card (Model CRL-G18U-P3DF, which is the base card that supports additional Edge TPU modules, up to 16x for the CRL-G116U-P3DF configuration).
• Documentation (this user manual).

4. Specifications

The following table details the technical specifications of the SmartFly info PCIe Gen3 AI Accelerator Card:

Figure 3: Official datasheet providing detailed specifications and a block diagram of the AI Accelerator Card.

Figure 4: Detailed specifications and an illustration of the advanced thermal design, including twin Axial-tech turbofans and a copper heatsink.

5. Setup

5.1. Hardware Installation

1. Prepare your system: Power off your computer and disconnect the power cable. Open the computer case.
2. Locate a PCIe slot: Identify an available PCI Express Gen 3 x16 slot on your motherboard.
3. Insert the card: Carefully align the AI Accelerator Card with the PCIe slot and press down firmly until it is securely seated. Ensure the retention clip (if present) locks into place.
4. Connect power: Connect a 6-pin 12V external power cable from your power supply unit (PSU) to the power connector on the AI Accelerator Card.
5. Secure the card: Fasten the card to the computer case with the appropriate screw.
6. Close the case: Replace the computer case cover and reconnect the power cable.

Figure 5: Rear view of the AI Accelerator Card, illustrating the PCIe connector for proper installation into a motherboard slot.

Figure 6: Side view of the AI Accelerator Card, showing the 6-pin external power connector that must be connected to the PSU.

5.2. Software Installation

The AI Accelerator Card requires specific software to function correctly. It is compatible with Linux-based operating systems.

1. Operating System: Ensure your system is running a supported operating system such as Ubuntu 18.04 or Debian 10. For the latest OS support list, please refer to the official ASUS IoT website.
2. TensorFlow Lite: The card is designed to run pre-trained ML models from Google TensorFlow Lite. Install the necessary TensorFlow Lite runtime and development tools.
3. Google Coral Tool Set: Utilize the comprehensive Google Coral Tool Set for compiling and deploying models. This includes tools for Edge TPU runtime, Python API, and model compilers.

6. Operating Instructions

The SmartFly info PCIe Gen3 AI Accelerator Card is optimized for Edge AI inference. Once installed and configured with the appropriate software, you can deploy and run your machine learning models.

• Model Deployment: Compile your pre-trained ML models (e.g., from Google TensorFlow Lite) for the Edge TPU using the provided tools.
• Application Integration: Integrate the compiled models into your applications. The card supports various frameworks and can be used with visual programming tools like ASUS IoT Node-RED AI Nodes for wiring AI applications.
• Inference Execution: Execute inference tasks on the card. The Edge TPU is designed for high-efficiency, low-power inference at the edge.
• Monitoring: Monitor the performance of your AI applications, including inference time, CPU, and RAM usage, through relevant system tools or dedicated dashboards.

Figure 7: An example of using ASUS IoT Node-RED AI Nodes for visual design of AI applications and monitoring inference results and performance metrics.

7. Maintenance

To ensure the longevity and optimal performance of your AI Accelerator Card, follow these maintenance guidelines:

• Cleaning: Periodically clean the card and its fans to prevent dust buildup, which can impede airflow and lead to overheating. Use compressed air to gently remove dust. Ensure the system is powered off and unplugged before cleaning.
• Ventilation: Ensure your computer case has adequate airflow. Do not obstruct the card's fans or heatsink.
• Temperature Monitoring: Monitor the operating temperature of your system, especially under heavy AI inference loads, to ensure it remains within the specified operating temperature range (0°C to 55°C).

8. Troubleshooting

If you encounter issues with your AI Accelerator Card, consider the following basic troubleshooting steps:

• No Detection: Ensure the card is fully seated in the PCIe slot and the 6-pin power connector is securely attached. Check your system's BIOS/UEFI settings to confirm PCIe slot detection.
• Performance Issues: Verify that the correct drivers and TensorFlow Lite runtime are installed. Check for any background processes consuming excessive system resources. Ensure adequate cooling.
• Software Errors: Consult the documentation for TensorFlow Lite and Google Coral tools for specific error messages. Ensure your operating system is up-to-date.
• System Instability: If your system crashes or becomes unstable, ensure your power supply unit (PSU) can provide sufficient power to all components, including the AI Accelerator Card.

If problems persist, refer to the support section for further assistance.

9. Warranty Information

Warranty information for the SmartFly info PCIe Gen3 AI Accelerator Card is not explicitly specified in the provided product details. Please refer to your purchase documentation or contact the seller/manufacturer directly for warranty terms and conditions.

10. Support

For technical assistance, additional resources, or to report issues, please use the following contact information:

• Wiki/Documentation: For comprehensive guides and information, visit the official product wiki: https://youyeetoo.com/blog/ai-accelerator-pcie-card-94
• Technical Support Email: For technical issues after purchase, you can contact the support team at: monica@smartfire.cn
• Seller Contact: Alternatively, you can click on "SmartFly Tech" on the product page and ask a question directly through the Amazon platform.