Projects
Projects
Embedded Files
2021-2022
Designed, assembled, configured, tested, and operated a compute platform that scaled to six NVIDIA RTX 3090 GPUs. The system used a single motherboard, PCIe risers, Windows 10, monitoring software, and two 1,300-watt power supplies.
Key result: Operated all six GPUs together while diagnosing thermal throttling, failed PCIe risers, disappearing devices, driver issues, unexpected shutdowns, and power-related failures.
Outcome: The platform required a 200-amp electrical-panel upgrade, dedicated power circuits, room-level cooling improvements, and continued monitoring to remain reliable.
Lesson: Complex systems require active testing, failure analysis, and operational ownership—not simply compatible components.
2022
Repurposed the six-GPU platform to evaluate local AI and machine-learning workloads. I removed the original mining configuration, installed GPT4All, enabled CUDA acceleration, and tested local language-model inference across the available RTX 3090 GPUs.
Key result: Evaluated local LLM operation, prompt workflows, GPU utilization, and agent-style experimentation while identifying the limitations of the existing system architecture.
Outcome: The PCIe riser-based design created data-transfer and communication bottlenecks that prevented the GPUs from functioning as one unified memory pool.
Lesson: High aggregate GPU capacity does not guarantee strong AI performance; system architecture must match the communication and memory requirements of the intended workload.
2022 - Present
Designed, assembled, configured, and tested dedicated workstations for local AI, image generation, software development, data analysis, content creation, and gaming. The systems repurposed RTX 3090 GPUs alongside new hardware, Windows 11, updated NVIDIA drivers, and CUDA-compatible software environments.
Key result: Delivered purpose-built systems tailored to different workloads, user requirements, performance expectations, and budgets.
Outcome: Completed multiple workstation configurations that were retained, sold, or gifted, while maintaining a separate system for continued personal experimentation and technical development.
Lesson: Reliable system design requires balancing performance, compatibility, thermal management, maintainability, cost, and the actual intended workload.
Page updated
Google Sites
Report abuse