Engineering Notes: Deploying A100 GPUs with MIG in OpenStack

This series of articles is a collection of engineering notes based on our hands-on research and testing with NVIDIA A100 GPUs, Multi-Instance GPU (MIG), and OpenStack integration. As part of OpenMetal’s open source ethos, we’re sharing these lessons learned to help the broader community solve real-world infrastructure challenges.

These articles are not intended to be a comprehensive deployment guide or cover every possible scenario. Instead, they document specific problems we encountered, how we addressed them, and the insights gained along the way. We hope they serve as a useful resource for engineers working through similar issues.

Customers of OpenMetal do not need to solve these problems themselves. Our GPU solutions are delivered ready-to-go, and we’re available to assist with advanced or non-standard GPU configurations as needed.

Whether you're building your own stack or just curious about how these systems work, we hope these notes are helpful.