Proxmox mini PC idle power: what the measured table shows and how to check your own node

Proxmox mini PC idle power is lower and more alike across models than most buyers expect. In Brandon Sheley's sourced table, most 1-liter PCs idle at about 9 to 16 watts at the wall, an Intel N100 box lands in the same range as a used OptiPlex Micro, and a workstation-class Minisforum MS-01 idles at about 25 to 29 watts.

The table, Mini PC Idle Power Consumption by Brandon Sheley (updated 19 October 2026), was just added to the awesome-proxmox-ve list. Its rule is simple: every row is a published wall-power measurement with a link to the review it came from. I opened each linked review to check the numbers below, then looked at what the Linux and Proxmox documentation says about measuring and lowering idle draw on your own node.

What the table measures

Each row gives a model, its CPU, an idle figure, a load figure and the reviewer. Sheley describes the figures as wall power on 120V, taken from CNX Software and ServeTheHome reviews. He also points out that the TDP printed on the box (6W, 15W, 35W) "tells you very little about idle draw."

These are the idle readings as the source reviews state them:

Model	CPU	Idle at the wall	Source
Beelink EQ12	Intel N100	12.1 to 14.3W	CNX Software
Beelink EQ12 Pro	Intel i3-N305	9 to 12W	ServeTheHome
HP EliteDesk 800 G3 Mini	Core i5-7500T	around 12W	ServeTheHome
HP ProDesk 600 G4 Mini	quad-core 8th gen	12 to 16W	ServeTheHome
HP EliteDesk 800 G6 Mini	Core i5-10500T	around 11 to 12W	ServeTheHome
Dell OptiPlex 7070 Micro	six-core 9th gen	just over 13W	ServeTheHome
Dell OptiPlex 7090 Micro	10th gen, 35W	around 12 to 14W	ServeTheHome
Lenovo ThinkCentre M920q	quad-core	12 to 15W	ServeTheHome
Lenovo ThinkCentre M720q	six-core Core i5	27W	ServeTheHome
Lenovo ThinkCentre M75q Gen 2	Ryzen 7 Pro 4750GE	around 12W	ServeTheHome
Minisforum MS-01	Core i9, 14 cores	around 25 to 29W	ServeTheHome

Sheley's page also lists load figures for each model. For the full set, and for his yearly cost estimates at the U.S. average electricity rate, read the original table.

How to read the numbers

Treat a difference of a watt or two as noise. Sheley says so on the page, because each reviewer used their own OS, power settings, RAM and storage. Several ServeTheHome reviews say the same thing in their own words: "We generally assume these nodes will use 9-12W idle," and a unit at 12 to 15W was "a bit higher."

The setup behind a number matters as much as the model. CNX Software measured the Beelink EQ12 at 12.1W in Windows and 13.9W in Ubuntu with one Ethernet cable, and 13.0W and 14.3W with two cables connected. ServeTheHome saw 11 to 12W on the EQ12 Pro in high-performance mode and 9 to 10W in power-saving mode in Windows 11 Pro.

The M720q row stands out at 27W. Sheley flags it himself: ServeTheHome's M920q, in the same chassis, idled at 12 to 15W, so the CPU and configuration of a single unit can move the result a long way. If one model in a comparison looks far off, read the source review before you rule the hardware out.

Workstation-class minis cost more to idle. ServeTheHome measured the MS-01 at around 25 to 29W idle and wrote that idle power "was higher than the average mini PC." You get 10GbE, a PCIe slot and more cores for that. Whether that is worth roughly double the idle draw depends on what the node runs.

Yearly energy is easy to work out from any idle number. Sheley's FAQ gives the rule: multiply the watts by 8.76 to get kWh per year, then multiply by your rate. A 13W node comes to about 114 kWh a year. For one node that is small. For a three-node cluster that never sleeps, it adds up, which is one reason to give a small cluster its third quorum vote from a low-power device.

A Proxmox data point from the N100 review

The CNX Software EQ12 review ran Proxmox with pfSense and TrueNAS VMs. In the comments under the review, its author Ian W. Morrison, who posts as Linuxium, reported that the CPU scaling governor in Proxmox was set to performance by default. Switching it to powersave took idle from 13.9W to 12.0W with one Ethernet cable on the original 5.15 kernel. After a kernel upgrade he reported 12.0W with performance and 10.8W with powersave.

That is one unit and one comment thread, so test the governor change on your own hardware before you count on the savings.

How to measure your own node

Use a meter at the wall for watts. Sheley recommends a plug-in meter for the most accurate single reading and a smart plug with energy monitoring to log draw over time, which gives you a real daily average. Measure with the drives, NICs and VMs you will really run, wait a few minutes after boot, then measure again under your normal load.

Powertop helps you find out where the power goes. The powertop README describes it as a tool to diagnose power consumption and power management issues, with Overview, Idle stats, Frequency stats, Device stats, Tunables and WakeUp tabs. It also says its per-device power estimates are built while the machine runs on battery, so on a mains-powered mini PC the wall meter gives the number you should trust. Install the powertop package on the Proxmox host and run it as root.

A few commands from the powertop README are useful on a server:

• powertop --html --time=60 measures for 60 seconds and writes an HTML report.
• powertop --auto-tune-dump prints the commands that --auto-tune would run, without applying them.
• powertop --auto-tune applies every recommended setting at once.

Run the dump first and read it. The ArchWiki Powertop page notes that interactive changes do not survive a reboot, shows a systemd unit to apply auto-tune at boot, and adds a line to stop plugged-in USB input devices from being put to sleep. On a headless node that matters less, but USB network adapters and disks deserve a check.

Settings that change idle draw

C-states in the BIOS come first. Sheley's advice for Proxmox is to enable the CPU's low-power states in the BIOS, turn off devices you do not use, and avoid passing a GPU to a VM that keeps it awake. The Idle stats tab in powertop shows whether the cores reach deep states at all.

The CPU governor is next. The kernel's CPU performance scaling documentation explains that the performance governor requests the highest frequency within the policy limit and powersave requests the lowest, and that with the intel_pstate driver the setting picks the driver's own scaling algorithm. Check what you have with cat /sys/devices/system/cpu/cpufreq/policy*/scaling_governor.

PCIe ASPM can matter for NICs and NVMe drives. The ArchWiki Power management page says the BIOS enables or disables ASPM at boot based on hardware support. You can check with lspci -vv | grep 'ASPM.*abled;' and read the current policy from /sys/module/pcie_aspm/parameters/policy. It also documents the pcie_aspm=force kernel parameter and warns that forcing ASPM on unsupported hardware can raise power use or cause freezes and kernel panics, so keep a way to remove the parameter before you reboot.

On Proxmox, kernel parameters go in one of two places. The Proxmox wiki Host Bootloader page says GRUB systems use GRUB_CMDLINE_LINUX_DEFAULT in /etc/default/grub followed by update-grub, and systemd-boot systems use one line in /etc/kernel/cmdline followed by proxmox-boot-tool refresh. Change one setting at a time and read the wall meter after each one, so you know which change did what.

If you are building a small cluster, plan the shutdown side too. A UPS that powers off nodes in the wrong order can do more damage than a few extra watts, as covered in Proxmox UPS shutdown and cluster order. For a real-world mix of cheap minis, see cheap Lenovo mini PCs and a Raspberry Pi 5 as a homelab cluster.