Running Hailo-8 with BeagleBone AI-64 — Software SDK/Driver Support?

General
1

Hello everyone,

I’m working on a project where I’d like to attach a Hailo-8 AI accelerator to the BeagleBone AI-64 through its M.2 E-key connector. Hardware-wise, I understand I may need an adapter and will be limited to a single PCIe lane, but my main concern right now is the software support.

  • Does the HailoRT / Hailo Dataflow Compiler officially support running on the BeagleBone AI-64 (Debian arm64)?

  • Has anyone here successfully installed the Hailo drivers and runtime on this board?

  • Are there prebuilt packages or instructions available for BeagleBone (similar to what exists for Raspberry Pi or x86 Ubuntu)?

  • Any tips on kernel version compatibility or pitfalls I should expect during installation?

My goal is to compile models with Hailo’s toolchain and deploy them directly on the BeagleBone AI-64. Before I dive into setting everything up, I’d like to confirm whether this software stack has been tested or if there are workarounds (e.g., cross-compiling, building from source, or using containerized environments).

Thanks in advance to anyone who can share their experience or point me toward the right documentation!

2

Hi @helter_skelter

The DataFlow compiler is our SDK to convert the AI models, and it is not meant for embedded systems. Instead, it must be installed on a PC with good CPU, 16+ / 32 (recommended) GB of RAM and ideally an NVIDIA GPU if you want to run some advanced optimization.
Only HailoRT and the PCIe driver are required on your embedded board.
The combo BeagleBone AI-64 with Debian is not in our validation infrastructure, but we have several customers that already integrated the Hailo-8 with the TDA4 SoC, so it should not be a problem.

Not on this board specifically, but definitely on the TD4VM EVK.

You can follow the HailoRT User Guide to install the driver and the HailoRT library. We provide either .DEB files - that you can download from our Developer Zone - or source code that you can compile yourself. If you plan to move to Yocto in the future, we also have meta-layers that you can integrate in your build.