Power consumption of routers and associated networking equipment [collection]

  • type: Raspberry Pi 4 (running HA)
  • device: Rpi4
  • energy/h: 12
  • min_power: 10W
  • max_power: 15.8W
  • energy: 0.29kWh
  • duration: 24hours
  • mainspower: 236V @50Hz
  • powermeter: Sonoff Dual R3
  • date: 26/07/2023
  • reporter: @KOA
  • details: Homeassitant 2023.7.3 with ice tower cooler with larger vertical fan. The fan was switched off and used a separate 4inch fan( not measured on the same port). Externally powered USB hub with generic RTL-SDR USB stick + ADATA 256GB SATA SSD connected directly to the RPI4.
1 Like

Hi KOA,
I added this to the tables above, please let me know if this needs changes. I was not sure about the type (which describes the function, like router, modem, whatever, and I am not sure whether homeassistent is a function), other than that, thanks a lot, the table becomes more useful with every addition...

it can be run in OpenWrt too.

1 Like

Just tested the following:

Raspberry Pi 3B (not B+)
Running as dumb AP only, with USB MT7610U 802.11ac 5GHz WiFi, with onboard 802.11n 2.4GHz temporarily ON, power consumption is only 2.2W

Another data point (though it might not be very useful since it's too old):

Raspberry Pi B Rev 2 (yes the first gen!), together with ASIX 100Mbps USB NIC (the one from Apple Mac), 2.5W idle, when I run wireguard test to put it on load the power consumption becomes 3W

I added my TP-Link EAP615-WALL to the list

    type: router
    device: Zyxel WSM20
    energy/h: 4.02 W
    min_power: n/a
    max_power: n/a
    energy: 0.579 kWh
    duration: 144 h
    mainspower: 230V @ 50 Hz
    powermeter: Voltcraft SEM 6000
    date: 202404
    reporter: @Luke02
    details: 23.05.2, dumb AP via WDS, firewall/odhcp/dnsmasq disabled, 0 ethernet ports connected, both radios enabled
• type: router
• device: Linksys-AE635v3
• energy: 4.96V (before 12V converter)
• min_power: 0.82W
• max_power: 1.97W
• energy: 16 mAh
• duration: 10 minutes
• powermeter: aliexpress noname USB
• date: 12/08/2024
• reporter: ear_connection
• details: router reboots shortly if I start downloading of a large file by wifi. It works fine with one ethernet. Probably, 5V 2A battery and 12V converter cant deliver enough power.

Hi everyone ,

TL;DR – We’re building a DC-powered industrial gateway around the 8Devices Habanero (Qualcomm IPQ4029/IPQ4019) module and are hitting a thermal ceiling. Because the module’s core rail is fixed 1.1 V, the usual Qualcomm HW-DVFS path (CONFIG_ARM_QCOM_CPUFREQ_HW) is disabled in the vendor BSP, so only frequency-scaling is left—and dropping all cores from 716 MHz to 48 MHz shaves barely 4 °C off the heatsink. We’d love to compare notes with anyone who has squeezed more watts out of an IPQ40xx board in OpenWrt.


Hardware / baseline · SoM: Habanero – dual-band Wi-Fi 5, 128 MB SPI-NAND, 512 MB DDR3L8devices

· Power rails per datasheet: 3.3 V (I/O), 1.35 V (MX rail), 1.1 V (CPU & CX rails) – all fixed regulators, no SPMI PMIC

· Wi-Fi MAC/PHY is QCA4019/ath10k; AP mode 24×7.

· Firmware: OpenWrt 23.05.0-rc3, r23389-5deed175a5 , Linux kernel 5.15.127, CONFIG_CPU_FREQ=y, OPP 48/200/500/716 MHz present; governor switched to ondemand.


What we’ve tried / observed

  1. CPU freq sweep (716 → 48 MHz) – negligible board-level power drop (makes sense: V² term is fixed)

  2. Hot-plugging 3 cores – similar result; workload is light but radios stay warm.

  3. wifi down / user “Wi-Fi OFF” toggle – interfaces disappear per /sbin/wifi helper, but maybe RF front-end still live.

  4. Checked .config – CONFIG_ARM_QCOM_CPUFREQ_HW & NVMEM not set, as in upstream ipq40xx patch-set.


Looking for help / ideas

· Clock-/power-gating tricks – Has anyone measured real savings from enabling deeper idle (CXSD/AOSD) on fixed-rail IPQ40xx? Any DTS patches to boost residency?

· DDR or bus down-clock – Is there a clean way to run DDR3L at 533 MT/s on this SoC? Datasheet hints timings go down to 48 MHz for SDC1.

· Wi-Fi power – Does ath10k ever cut the PA supply in AP mode? CT firmware or OpenWrt hacks welcome.

· clk_ignore_unused – Did you gain anything by clearing that flag (let CCF gate more clocks)?

· Light hardware tweak – Anyone here lowered the Habanero’s CPU rail by resistor swap and kept 716 MHz stable? Results?

· Any other oddball knobs (PCIe L1 sub-states, USB autosuspend, etc.) that moved the needle?

We’re happy to share logs, DTS fragments and thermal meassurements. End-goal is a short cost/benefit report before we decide whether to respin hardware or stay software-only.

Thanks in advance for any pointers, patches or “don’t bother, tried that!” stories.

— Jesper & team