Building a compute fleet from incompatible parts.
The project involved building and operating a multi-rig compute fleet in the late 2010s for cryptocurrency mining and related experiments. Inconsistent hardware availability made component sourcing, compatibility, and configuration recurring engineering challenges.
As the fleet's builder and day-to-day technical operator, I had to make mismatched hardware and facility constraints work together as a reliable operation.
Hardware integration crossed several disciplines.
The installation grew to approximately 12–13 open-air GPU rigs. It used configurations built around GeForce GTX 1080, Radeon Vega, and Radeon RX-series cards, along with varied motherboards and CPUs selected around available GPU-slot and platform requirements.
Custom physical builds.
Fabricated open-air frames, arranged GPUs using PCIe risers/extensions, and adapted motherboard and multi-PSU configurations to differing card densities.
Power and cable planning.
Coordinated capacity expansion with an electrician who performed electrical work; organized equipment placement, cable runs, and drop-ceiling routing.
Managing continuous heat.
Dealt with room temperatures near 80°F through multiple stand-alone AC units and distributed thermal-management measures.
The fleet came together while I was completing my computer science education.
Making the systems operable.
Linux administration became central to the project. Driver compatibility, GPU recognition, and stability issues demanded documentation, experimentation, and persistent troubleshooting.
- Built shared keyboard, mouse, and display-switching arrangements to reduce the friction of managing many machines locally.
- Established remote monitoring and management as practical within the constraints of the evolving fleet.
- Handled hardware incompatibilities, GPU recognition and driver issues, stability problems, and temperature management.
- Assessed the risks of later physical relocation and considered possible alternative business uses for the installed computing capacity.
The practical lesson was clear: capable components become a reliable operation only when power, cooling, access, software compatibility, and recovery are engineered together.
A facilities-driven closeout.
The operation was ultimately discontinued after the property housing the fleet was sold. Its electrical capacity, thermal load, and physical footprint made relocation dependent on finding another suitable facility; no replacement space was available. The subsequent COVID-19 disruption ended the prospect of restarting the project.
Roles and responsibilities.
Facility electrical work was handled by a qualified electrician; hardware integration, configuration, and daily operation of the fleet were my responsibility.
What running the fleet taught me.
Operating the fleet reinforced a systems view of hardware, operating systems, facilities, and everyday reliability—a perspective that continues to inform infrastructure and integration work.
It also reinforced the need to evaluate whether a technically workable system makes sense to operate over time.