Methodology & sources

The manufacturer-UNPUBLISHED deciding facts for picking a homelab box: MEASURED idle watts (wall power, not TDP), ECC support, NIC chipset, RAM ceiling, IOMMU/passthrough behaviour, and AMD reset-bug status. Every deciding field carries a source citation and a date, or is omitted. Spec sheets list TDP — explicitly the wrong metric for an always-on box. The third-party measurement is the only source.

The rule: cited or omitted

Every deciding field for every model is one of two things: a value with a dated third-party source link, or blank. There is no third option. We never assert a number we can't cite, because a wrong idle-watt or passthrough claim costs a buyer a $200–800 mistake. A machine-checked test in the repo fails the build if any present field is missing its citation.

Why TDP is excluded

Manufacturer spec sheets list TDP — a thermal-design figure for sizing a cooler, not a power-draw figure. The same chassis with a 35 W-TDP "T" chip vs a 65 W-TDP desktop chip idles ~11 W vs ~18 W. The only useful number for an always-on box is the wall-measured idle draw, which no manufacturer publishes. So every idle figure here comes from a third party who put a meter on the wall — and we note the method (Kill-A-Watt, smart-plug week average, iDRAC) and that idle varies with OS, 120 V vs 240 V mains, and powertop tuning.

Fields we track

• Measured idle watts — wall power, with method + source.
• ECC support — "claimed/unverified" is stated honestly where the community never validated a working ECC module.
• NIC chipset — i226-V (good) vs i225-V (dropout bug) vs Realtek (Linux-only) vs Intel I219 (rock-solid). Often missing from spec sheets.
• RAM ceiling — real maximum, soldered vs SODIMM.
• IOMMU / passthrough — community reports of group cleanliness, iGPU passthrough difficulty, SR-IOV vGPU.
• AMD reset bug — flagged on AMD-APU models with the vendor-reset workaround source.
• Proxmox fit — our editorial synthesis derived from the cited fields above; clearly labelled as a verdict, not a spec.

All sources (56)

Open data

The full dataset (51 models) is public and reusable on GitHub (JSON + CSV). Corrections and new measurements are welcome as PRs — the point is to aggregate the community's scattered wall-meter readings into one queryable place.

← Back to the finder