Builds
Hardware builds across different use cases — high-end homelab AI rigs, budget inference machines, and server platforms.
Published April 10, 2026
Builds
Theory only goes so far. These are the actual machines — the specs, the PCIe topology, the constraints, and what each build is used for.
Not every build is optimal. Some are budget-constrained, some are experiments, one is a dual-socket server that is still being assembled. The point is to document real configurations and what they mean in practice — not ideal specs on paper.
In This Section
- X570 · Ryzen 9 5950X · Dual RTX 3090 — the primary homelab rig. High core count, 48 GB VRAM across two GPUs, PCIe 4.0.
- Supermicro X10DRC-LN4 · Xeon E5-2696 V4 — a dual-socket server platform being built out for compute and VM workloads.
- i3-8100 · GTX 1060 6 GB — the floor. A minimal AI inference machine that documents what is possible on constrained hardware.
What These Pages Cover
Each build page covers the PCIe topology — which devices are CPU-direct, what shares chipset lanes, and where the bandwidth constraints actually are. That is the thing most build guides skip.
The specs alone do not tell you whether your second NVMe slot competes with your GPU, whether a dual-GPU setup is a true bifurcation or one card running at x4, or why your Thunderbolt port tops out lower than expected. The topology does.