DIY Udoo Quad Distributed Computing Cluster

Overview

This project is a DIY Distributed Computing Cluster built from 13 Udoo Quad SBCs, running Armbian Linux, and tied together with custom 3D-printed enclosures, a cooling manifold, and a custom-built power distribution board.

✨ Read the full blog post here: Building a Distributed Computing Cluster with Udoo Quads

Features

✔ 13-node Parallel Computing Cluster
✔ Armbian 21.08.5 (XFCE) running on each node
✔ MPICH + MPI4Py installed for distributed Python programs
✔ Custom-built power distribution board for compact wiring
✔ 3D-printed modular frame with airflow-optimized cooling
✔ Centralized controller with Python GUI management

Files

• main.py: main GUI file

• md5_attack.py: MD5 cracker using MPI

• machinefile: list of all IP addresses on the clusters local network

• Due to the large filesize (>1GB), full-assembly.stl is not included in /stl. Please convert from the substantially smaller full-assembly.step file.

Cluster Overview

Hardware Components

• 13x Udoo Quad Single Board Computers
• 2x 8-port TRENDnet TEG-S80g Gigabit Switches
• 300W 12V Power Supply
• 120mm Cooling Fan + Custom 3D-printed Manifold
• Custom-made Power Distribution Board
• 14x Ethernet Cables (~12 inches each)
• 13x 8/16GB MicroSD Cards
• Wire, Connectors, JST and Molex Pins, Protoboard

Software & Code

The cluster uses Armbian Linux running MPICH for parallel message passing and MPI4Py for Python compatibility. Custom scripts allow:

• Parallel execution of Python programs
• GUI-based board management (status, SSH access, file transfers)
• Automatic SSH key setup for passwordless communication

Example MPI Test

mpiexec -f machinefile -n 13 hostname

Python GUI Control Panel

• Monitor board temperatures and uptime
• Push code updates globally
• Open SSH terminals from a simple interface

3D Printing Files & PCB Design

All STL, STEP, and library are provided in this repository.

🖌️ Onshape CAD Models: Onshape Project

Setup

1. Flash Armbian 21.08.5 XFCE to each SD card.
2. Install MPICH and MPI4Py.
3. Setup SSH key exchange between all nodes.
4. Define machinefile listing all IP addresses.
5. Launch programs across the cluster using mpiexec.

Networking

• Nodes interconnected via two 8-port switches.
• Static IPs assigned for easier configuration.

Power Distribution

• 300W 12V PSU feeds each board through custom-made JST connectors.
• All boards mounted with 3D-printed supports.

Known Limitations & Future Improvements

⚠ Performance Gains Limited — Only ~2x speedup observed; bottlenecked by network overhead and Python interpreter.
⚠ Hardware Fragility — Two boards damaged due to passive overheating.
⚠ Control Circuit Issues — Initial design overheated voltage regulators; needs proper buck converters.

Future Ideas

• Upgrade to Kubernetes or Docker Swarm.
• Stronger 3D-printed base frame.
• Improved software parallelism (cryptomining, matrix ops).

License

This project is open-source under the MIT License.

Author

Developed by Ballistyxx
📧 Contact: GitHub Issues

---

✨ If you like this project, consider giving it a ⭐ on GitHub!