The original goal was to build a scalable mining platform that could grow as GPUs became available.

Rather than purchasing all of the hardware at once, I expanded the system gradually. Each additional GPU increased the system's earning potential, but also increased:

• Electrical demand

• Thermal output

• Cooling requirements

• Hardware complexity

• Monitoring needs

• Risk of system instability

What began as a computer-building project eventually became a small infrastructure and operations project.

At its maximum scale, the platform included:

• Six NVIDIA RTX 3090 GPUs

• A single motherboard

• PCIe riser connections

• Two 1,300-watt power supplies

• 128 GB SATA SSD

• Windows 10

• NiceHash

• HWMonitor

• Additional ventilation and cooling equipment

Debugging Process

When a failure occurred, I worked to reduce the number of possible causes rather than changing multiple variables at once.

Depending on the problem, the process included:

• Identifying the affected GPU or subsystem

• Checking physical connections and power delivery

• Inspecting the PCIe riser and associated cabling

• Confirming whether the device was recognized by Windows

• Reviewing temperatures and operating behavior

• Reinstalling or adjusting drivers and software settings

• Testing the system again under load

• Monitoring whether the failure returned

• Documenting the result and determining the next corrective action

This was an independent project rather than a formal enterprise qualification laboratory, but it required many of the same core disciplines: controlled testing, component isolation, failure reproduction, corrective action, retesting, and continued monitoring.

I managed the project from research through operation, including:

• Researching components and system requirements

• Sourcing and assembling hardware

• Installing and configuring Windows

• Installing and maintaining NVIDIA drivers

• Configuring NiceHash and monitoring software

• Tuning GPU settings

• Monitoring temperatures and system behavior

• Investigating recurring hardware and software failures

• Planning airflow and cooling improvements

• Determining electrical requirements

• Coordinating electrical upgrades with a licensed electrician

• Deciding when continued operation was no longer worthwhile

As the platform expanded, thermal output increased significantly.

The system initially operated inside the home, but the surrounding room became increasingly difficult to cool. GPUs would throttle when temperatures rose, reducing performance and increasing the risk of shutdowns.

• Increased spacing between GPUs

• Adjusted GPU fan curves

• Improved room ventilation

• Installed an exhaust fan

• Added air conditioning

• Relocated the platform to a more isolated operating area

Design decision: Its heat output required room-level environmental planning rather than component-level cooling alone.

The six-GPU platform placed demands on the home's electrical system that exceeded the original operating setup.

After evaluating the system's requirements, I coordinated upgrades with a licensed electrician.

Infrastructure Changes

• Replaced the existing 150-amp electrical panel with a 42-space, 200-amp panel

• Added a dedicated 30-amp, 240-volt circuit for the GPU platform

• Added a separate 15-amp, 120-volt circuit for monitoring, lighting, and supporting equipment

• Completed required bonding and code-related upgrades

My role: I determined the system's operational requirements and coordinated the work. The electrical installation itself was completed by a licensed professional.

The platform reached its full six-GPU configuration and operated with all six GPUs together for approximately three months.

At the time, the completed system demonstrated an estimated annualized earning potential of approximately 2.68  BTC, depending on market conditions, mining difficulty, power costs, and system availability.

The project also resulted in:

• A completed six-GPU operating environment

• A major residential electrical upgrade

• Dedicated power circuits

• Improved cooling and ventilation

• Practical experience with GPU monitoring and tuning

• Extensive troubleshooting experience

• A clearer understanding of infrastructure operating costs

Qualification and Reliability Outcomes

• Confirmed that six GPUs could operate concurrently in the completed configuration

• Identified PCIe risers as a recurring point of failure

• Distinguished thermal failures from connectivity, power, and driver-related failures

• Verified improved operating conditions after airflow, cooling, and electrical changes

• Developed a repeatable component-isolation and retesting process

• Documented limitations that later influenced the decision to repurpose the hardware

• Applied lessons from the platform to more maintainable workstation designs

The system was technically functional, but continued operation required more maintenance, cooling, monitoring, and electrical support than I originally expected.

The decision to stop was based on the complete operating picture:

• Electricity consumption

• Cooling requirements

• Hardware wear

• Market variability

• Recurring maintenance

• Time spent monitoring and troubleshooting

• Limited predictability of long-term returns

Ending the project was not the result of a single system failure. It was an operational decision based on whether the platform remained worthwhile to maintain.

After mining operations ended, I investigated whether the same six-GPU hardware could be repurposed for local artificial-intelligence workloads.

Although the system contained substantial total GPU memory, its PCIe riser-based architecture had been designed for workloads in which GPUs operated mostly independently.

Local AI workloads introduced different requirements, particularly when data needed to move between GPUs.

This led to the next project:

Local AI and Machine Learning Experimentation →

Large amounts of hardware do not automatically create an effective system.

The final result depends on how well the architecture matches the workload, including:

• Power delivery

• Cooling

• Interconnect bandwidth

• Software support

• Reliability

• Operating cost

• Maintenance requirements

The project strengthened my understanding that technical systems require continuous evaluation—not only during design and construction, but throughout their operating life.

The platform required regular monitoring and intervention.

Recurring problems included:

• GPU overheating

• GPU throttling

• GPUs disappearing from the operating system

• Failed PCIe risers

• Breaker trips

• Driver and NiceHash configuration issues

• Unexpected system shutdowns

Troubleshooting often involved isolating individual components, checking power and riser connections, reinstalling drivers, adjusting settings, and observing the system under load.

Operational lesson: A system described as producing passive income still required active technical ownership, maintenance, and decision-making.