A mini-PC that a review measured at 8W idle can easily draw 25W in your rack, and the difference is almost never the hardware - it is BIOS defaults and an untuned OS. On a box that runs 24/7, closing that gap is the single highest-return tweak you can make: 15W saved is real money every year. Here is where it hides.

First, measure it

You cannot tune what you cannot see. A cheap smart plug with power reporting (or a Kill-A-Watt) at the wall is enough. Note the idle draw with the OS at rest, no VMs doing work. That is your baseline - and the number the power calculator turns into an annual cost. Re-measure after each change so you know what actually helped.

The big one: CPU C-states

Idle power is mostly about how deeply the CPU is allowed to sleep between tasks. Deep package C-states (C8/C10) can drop a modern Intel chip to 1-2W - but a single setting or device can pin the whole package at C2/C3 and cost you 10W+.

  • In the BIOS, enable C-states / "Package C-state limit = Auto" (or C10). Some vendors ship with C-states disabled for stability.
  • On Linux, install and run powertop. Its "Idle stats" tab shows how much time the package spends in each C-state. If it never reaches PC8+, something is holding it awake.
  • The usual culprit is the OS or a device generating constant interrupts. powertop --auto-tune applies the safe power-saving toggles; watch the wall meter as you do.

ASPM: the setting reviewers forget

Active State Power Management lets PCIe devices - your NVMe SSD, your NIC - drop into low-power link states when idle. If ASPM is off (a very common default), those devices keep the PCIe link, and often the whole CPU package, awake.

  • Enable ASPM in the BIOS (set to "Auto" or "L1").
  • On Linux, pcie_aspm=force on the kernel command line can help boards that ignore the BIOS setting - but test stability; a few NVMe drives misbehave with aggressive L1.

Getting C-states and ASPM right is usually 80% of the win, and takes a box from ~20W to under 10W.

The device culprits

Culprit Idle cost Fix
10GBASE-T copper NIC 3-5W per port Prefer SFP+ (see the 2.5GbE vs 10GbE guide)
Spinning HDDs 4-8W each Spin-down for cold data; SSDs for hot
RGB / bright status LEDs 1-3W Disable in BIOS/software
An always-100% background process 5-15W Find it: powertop, htop - a busy-loop stops C-states entirely
Discrete GPU with no idle management 10-15W Ensure the driver powers it down; passthrough VMs can pin it awake

The single biggest surprise is usually a 10G copper NIC or a runaway process holding the CPU out of its sleep states.

What to leave alone

  • Do not disable turbo or undervolt for idle - idle draw is about sleep states, not peak clocks; turbo costs nothing at rest.
  • Do not chase the last watt on a box that does real work. If your server is genuinely busy, idle tuning matters less than picking an efficient chip in the first place - see the best mini PC for Proxmox and the value frontier.
  • Stability first. If a deep C-state or forced ASPM causes lockups or NIC drops, back it off one notch. A box that stays up at 12W beats one that crashes at 8W.

A realistic target

For an N100/N150 box: 6-10W idle is achievable and normal once tuned. For a 6-8 core AMD or Intel box: 10-18W. For anything with a 10G copper NIC, spinning disks, or a discrete GPU, add their fixed cost on top - and that is exactly the cost the power calculator is there to show you before you buy.