Architecture
Choose boundaries between routing, storage, compute, management, and user-facing services.
FG / 01 / Field guide
Build the quiet, recoverable backbone for local AI, storage, networking, and the services worth owning.
Current briefing
Standing watch
Watch for quiet, low-power options that improve the next rack revision without creating a management or compatibility tax.
Open the brief →Primary-source watch
Track dated release notes that materially affect backup, clustering, storage, recovery, or hardware support.
Open the brief →Build watch
Collect practical builds that disclose power, noise, thermals, expansion limits, software choices, and failure modes.
Open the brief →Current briefing
An FCC Covered List update blocks regulatory authorization for new foreign-made consumer routers, raising boundary questions for modular edge hardware and home lab firewall appliances.
Open the brief →Current briefing
Intel Arc Pro B60 and B70 GPUs enable subscription-free SR-IOV virtualization on Proxmox VE, offering home labs and small enterprises a low-cost alternative to recurring Nvidia vGPU licensing.
Open the brief →Current briefing
Intel Arc Pro B60 and B70 workstation GPUs can be partitioned into independent virtual functions on Proxmox VE without recurring licensing fees, offering an accessible alternative to Nvidia vGPU stacks.
Open the brief →Current briefing
Demonstrations of Intel Arc Pro B60 and B70 workstation GPUs show functional SR-IOV virtualization in Proxmox VE, offering dense multi-user GPU slicing without recurring enterprise license fees.
Open the brief →Get your bearings
Treat the lab as a system with five layers: network, compute, storage, applications, and operations. The rack is not the product. Reliable services, controlled failure, and understandable recovery are the product.
Start here / Three moves
Inventory hosts, storage pools, network dependencies, power, backups, remote access, and the services that other people depend on.
Write the target architecture, data classes, backup path, validation checks, cutover order, rollback plan, and acceptable downtime before moving anything.
Record idle power, noise, temperatures, update cadence, alerts, credential handling, and one tested restore.
Build capability
Choose boundaries between routing, storage, compute, management, and user-facing services.
Patch, observe, back up, restore, document, and retire systems without depending on memory.
Match models and accelerators to real workloads, then account for thermals, power, memory, and data sensitivity.
Choose gear / Decision rule
Choose by workload, consequence of failure, idle power, noise, expansion, software compatibility, recoverability, and full-system cost—in that order. A spec sheet without an operating plan is not a recommendation.
Your source map
Reference layer
Original editorial work
A useful home lab is measured by recoverability and capability—not the amount of equipment in it.
Read the working note →Architecture, validation, rollback, and the operating baseline that should exist before the first file moves.
Read the working note →Pre-configured tiling environments promise to eliminate Linux configuration fatigue, but curated defaults cannot fully bypass the realities of rolling-release workstation maintenance.
Read the working note →Dedicated neural accelerators are finally arriving in compact homelab hardware, but their true utility hinges on runtime stability and pipeline offloading rather than raw throughput.
Read the working note →Recommendation ledger
Candidate
Open the product record →Each recommendation will include drawbacks, alternatives, last-reviewed date, evidence level, and affiliate disclosure.