Power consumption of routers and associated networking equipment [collection]

moeller0 · September 12, 2022, 9:03pm

This thread is intended to collect information about power consumption by OpenWrt routers as well as associated network equipment. Depending on the measurement devices available anything from immediate power (Watt) to energy (Joule or Watt Second) is fair game.

NOTE: This post should be editable by all, please feel free to add your own numbers to the table

To make a start (? denotes "missing"/not measured values):

type         ; device                   ;    energy/h;   min_power;   max_power;      energy;    duration;  mainspower;  powermeter;  date  ; reporter      ;details
             ;                          ;      [Wh/h];         [W];         [W];       [kWh];         [h];      [V@Hz];         [#];        ;               ;
modem        ; BT HomeHub 5A            ;        6.89;         2.2;         7.5;       0.376;       54.55;      230@50;           1; 202209 ; @moeller0     ;"OpenWrt r18441-ba7cee05ed, bridged-modem, WiFi disabled, VDSL2@PTM with vectoring and G.INP, 116.7/37 Mbps sync"
router       ; Turris Omnia             ;        7.98;         4.0;        12.0;       0.187;       23.43;      230@50;           1; 202209 ; @moeller0     ;"Turris OS 5.4.2, based on OpenWrt 19 dual core ARM a9 mvebu @1.6GHz, 2 radios 2.4GHz & 5GHz, 3 devices connected to switch, SQM/cake on wan 105/36 Mbps"
router       ; Mikrotik RB5009UPr+S+IN  ;        8.51;         8.2;        10.3;       1.2  ;      141   ;      230@50;           ?; 202209 ; @Arie         ;"OpenWRT 23.03, dual gigabit WAN, one gigabit LAN port connected."
router       ; Custom-built x86         ;           ?;         9.0;        20.0;           ?;           ?;      230@50;           3; 202209 ; @pico         ;"OpenWRT 22.03rc5, Pentium Gold G5600(8.Gen Coffee Lake 2+2HT 4GHz), Asrock B365 Pro4F, 1x8GB RAM, USB-Stick for OpenWRT, 1TB Corsair P510 for ksmbd, Broadcom PCIe 1GBit Lan x4 (74Y4064), Leike PSU (on German 230V)+Pico150 PSU, no WiFi, no input devices, no undervolt, no SQM so far, 1 LAN-device, ""max value"" = when routing 900MBit/s IPv6 (not tested: CPU burn-in likely ~50-55W)"
VoIP-base    ; Gigaset C610-IPA         ;        0.98;           ?;         1.4;       0.038;       38.73;      230@50;           1; 202209 ; @moeller0     ;"VoIP-basestation with a single DECT handset connected, used rarely"
NTP-server   ; Raspberry 4B             ;        2.00;       ?2.9?;         4.6;       0.080;       39.55;      230@50;           1; 202209 ; @moeller0     ;"Raspberry Pi 4B 4GB ram, with Uputronics Raspberry Pi GPS/RTC Expansion Board, operated as GPS disciplined NTP-server; odd readings with W readings not matching the aggregate energy usage"
switch       ; Zyxel GS1900-8HP         ;        4.99;         4.0;         6.2;       0.688;      137.83;      230@50;           1; 202209 ; @moeller0     ;"8-Port managed PoE switch; Zyxel firmware (V2.60(AAHI.3) | 01/05/2021); 5 ports connected, 0 devices powered over PoE"
router       ; TP-Link C7 v2            ;        6.28;         6  ;         7  ;       0.29 ;       47   ;      230@50;           4; 202209 ; @fodiator     ;"OpenWRT 22.03rc6, primarily AP with dynamic VLAN and 802.1x, dual band. No DHCP, DNS, USB or wired clients"
router       ; FriendlyElec ZeroPi      ;        1.39;         1.3;         1.8;       0.375;      269.6 ;      230@50;           5; 202209 ; @Barney       ;"FriendlyElec ZeroPi with TP-Link UE300; Download 50 MBit/s, upload 10 MBit/s, OpenWRT 21.02.0, 10-20 devices"
router       ; Belkin RT3200            ;        4.35;           ?;           ?;       5.01 ;     1152   ;      120@60;           6; 202211 ; @efahl        ;"Belkin RT3200, OpenWrt SNAPSHOT updated every Monday and rebooted, 2x1Gbe connections, 5G WiFi only, 2.4 radio turned off"
router       ; Custom-built x86         ;       16.45;	       3.8;        22.3;       2.785;	   169.3 ;	    230@50;	          7; 202211 ; @thejoker8814 ;"OpenWRT 21.02.3, Dual band Wifi 4/5, VDSL modem, Mitac PD10-RI, 4GB LPDDR3, 64GB SSD, (DNS, DHCP, NTP, BanIP, SQM, VPN), all measurements during normal operations (3-6 clients, mixed Wifi/ LAN)"
modem        ; Fritzbox 7520            ;        5.94;	       2.6;         7.0;       1.034;	   174.0 ;	    230@50;	          1; 202304 ; @moeller0     ;"OpenWrt SNAPSHOT, r22256-14334c222e, VDSL2 modem (VRX518) bridged to , WiFi: 2.4GHz disabled, 5GHz active, VDSL2@PTM with vectoring and G.INP, 116.7/37 Mbps sync"
homeassistant; Raspberry Pi 4           ;       12.00;	      10.0;        15.8;       0.29 ;	    24   ;	    236@50;	          8; 202307 ; @KOA          ;"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."
router       ; CWWK N5105 x86           ;        8.80;           ?;           ?;        lots;       2000+;      120@60;           6; 202310 ; @efahl        ;"OpenWrt SNAPHOT updated weekly, 4x2.5GbE with 2 plugged in, 8GB DDR4, WD Blue SN550 250GB NVMe SSD"
accesspoint  ; TP-Link EAP615-WALL      ;           ?;         3.9;         7.5;           ?;           ?;      50-57v;           ?; 202401 ; @ed8          ;"OpenWrt 23.05.2, measuring comes from the PoE switch

Column description:

type: what kind of device, like router, modem, switch, AP, more details go into the details column
device: short identifier of the device under test
energy/h: average power usage [Wh/h -> W]
min_power: minimal power draw during measurement period [W]
max_power: maximal power draw during measurement period [W]
energy: total energy consumed during the measurement in kilo Watt hours [kWh]
duration: duration of the test in hours with decimal fraction (so not hours.minutes)
mainspower: information about the AC mains power, specifically nominal voltage and frequency [V@Hz]
powermeter: code of the used powermeter according to the table below
date: concatenated year and month of measurement (e.g. 202209 for September 2022)
reporter: forum name of who reported the data
details: any details worth reporting that might be relevant for the power consumption, e.g. software version, which/whether radios were enabled, how many devices where connected

If any column should be missing or be ill defined, please let me know, always happy to add/correct columns.

power meter device codes [powermeter]:
#       model name
1       LogiLink EM0003 v.2
2       Voltcraft 4000
3       Perel 305EM5-G/TS electronic 45-25111
4       Brennenstuhl PM 231 E
5       Globaltronics GT-PM-04 (compatible with Logilink EM003)
6       Poniie PN1500
7       Voltcraft SEM 6500
8       Sonoff Dual R3

Initial measurements were taken with a LogiLink EM0003 v.2 energy meter that according to a July 30th 2022 article in German c't magazine has a deviation against a high quality reference of <= 5% (for loads in the 1-20 W range) and measures energy and real/active power.

This was motivated by discussions in other threads about the relative merits of using consolidated all-in-one WiFi-routers versus de-centralized solutions with dedicated devices for typical functions, like wired-only router, AP(s), managed switch, modem, NAS?... The idea is to simply collect energy use information that can be used as additional dimension when designing/deciding about a home-network's device zoo.

Changes:
20220917: reordered columns, to make the most relevant information fits into view without requiring horizontal scrolling. Also added ";" as column separators to make CSV import simpler.

notgood · September 14, 2022, 10:19am

Most of the consumer level measurement devices, i.e. AC socket power meters, are unbelievably inaccurate at sub 10W levels. Some are cheap enough to only measure apparent power in VA, instead of real power in Watts, completely ignoring Cos Phi power factor. So it kind of makes sense to post info about your measuring technique as well.

IMHO in case of 24h connected home network appliances, the only number that really matters is idle power usage.

A few numbers for random networking gear, idle power:
Netgear R7800, OpenWRT 22.03, both radios on, 9.3W
TP-Link TL-SG108PE, 3 non-PoE ports up at 1GBit/s, 3.5W
D-Link DGS-1005D I1, 2 ports up at 1Gbit/s, 1.3W

Measured with Voltcraft 4000 with a following accuracy:
1% at 5+ Watts
5% at 2-5 Watts
15% at <2 Watts.

moeller0 · September 14, 2022, 10:53am

I consider "long term average consumption will match idle power times duration" a hypothesis that can be easily tested, simply by collecting and averaging longer term data (say at least 24 hours to cover all usage periods). Especially for devices without a priori information about the idle power usage I see no way around measuring it. I added the min and max power readings (with all their issues, the power meter averages somewhat and does not give instantaneous power) to give a feel where in the continuum from min to max the average sits.

Well the LogiLink EM0003 v.2 I used was just tested by a reputable German computer magazine to have measurement errors <= 5% under low power conditions (20-1 Watt, but complete failure for a 0.1W load) average deviation from high quality references was at 2%. While not usably for a scientific study, I think that errors <= 5% are totally fine for the purpose of this thread (but then I opened this thread). But since you seem to know more about the details, please correct any misconceptions in the above.

Arie · September 14, 2022, 11:13am

energy      duration    energy/h    min_power   max_power   device                      details
 [kWh]           [h]      [Wh/h]          [W]         [W]
   1.2           141        8.51          8.2        10.3   Mikrotik RB5009UPr+S+IN     OpenWRT 23.03, dual gigabit WAN, one gigabit LAN port connected.

notgood · September 14, 2022, 11:15am

Of course, as long it is real/active power, 5% accuracy is totally fine.

Here is a recent graph, power consumed by router ~10W, switch ~4W, NAS Celeron J1900 with 4 sleeping HDD ~17W, all going through UPS.

Whole setup consumes ~32 Watts running 24/365.
Assuming electricity cost of 50 eurocents/KWh, thats at least 140 EUR yearly

moeller0 · September 14, 2022, 11:39am

Question: would everybody here be OK if I go and collect all entries into a single table (probably as @tmomas proposed over in the wiki with a link to this thread).

If yes, which information columns are you missing. maybe:
measurement device; year and month; connected devices, number of active radios; ...

My goal is to keep this simple, but if it can be made more useful easily I am happy to do so.

moeller0 · September 14, 2022, 11:53am

Gentlemen, I was off, we do not need to go over to the openWrt wiki, instead, the top post should now be editable.

Please feel free to add your data to the existing table, I will add a reporter column and a year/month column.

Pico · September 14, 2022, 8:37pm

I unfortunately did not succeed to edit the first post.

I suggest, to add another tag column for better comparability: the label name of the power meter used for measuring. You do need a reasonably good modern power meter for that. Most older ones are just good for measuring fridges and classic light bulbs, but have difficulties with LED bulbs and devices lower that 20W or show fantasy values, when trying to measure modern electronic switching power supplies (German: Schaltnetzteile).

x86 custom router with 22.03rc5:
Pentium Gold G5600(8Gen Coffee Lake 2+2HT 4GHz), Asrock B365 Pro4F, 1x8GB RAM, USB-Stick for OpenWRT, 1TB Corsair P510 for ksmbd, Broadcom PCIe 1GBit Lan x4 (74Y4064), Leike PSU (on German 230V)+Pico150 PSU, no WiFi, no input devices, no undervolt

9Watt booted openWRT idle (I’ve rounded upwards, reality is even a bit less)
13W when NATing ~400MBit/s IPv4 NAT (my ISP for unknown reasons won‘t go higher with IPv4)
20W during a 900MBit/s IPv6 Cable Internet Speedtest
I guess, it might draw around 50Watt, if I would go for a prime95 burn-in-test.
I have not configured or tested SQM so far
all values measured with 1 lan device, adding +1 LAN device ups power by + ~0.3W

I am really happy with that, as it allowed me to kick out my old hatred QNAP 119 NAS which also had 10-20W. Compared to the cost of a 1-4bay NAS, x86 parts aren‘t that much more expensive and additionally do cover routing.

Measured with: Perel 305EM5-G (rebrand of TS electronic 45-25111, which was the winner of German ct Magazine power meter measure shootout in 2021).

moeller0 · September 14, 2022, 8:54pm

Crap. @tmomas do you know why by chance?
@Pico if you format your data so it matches the above table I will happily add it while we figure out how to allow editing for all..

tmomas · September 14, 2022, 8:58pm

Technically it should be possible (editing wiki posts is allowed for TL1 and above, which is true for Pico).

Without knowing the details ("It didn't not work" :)), maybe it should be mentiond that wiki postings can be edited by clicking on the symbol marked in red:

Pico · September 14, 2022, 9:04pm

my pardon, layer8 problem, I missed to click the „edit wiki“ button afterwards. Looks doable now, will try.
I have no consumption over time values, I will just leave those empty for now.

moeller0 · September 14, 2022, 9:18pm

Yes, please do, shall we add an avg_power column?

Actually the Energy/h column is already an average power column as watt * h / h = watt, maybe I should simply rename it.

Pico · September 14, 2022, 9:32pm

To me, the energy/h column looks sufficient for that.

For my x86 I've added a comment at the end of the long line that the mentioned 20W was while routing 900mbit (as otherwise my x86-values would be unrealistic, as although a x86 has more CPU power, it will also draw a lot more power, if the CPU really gets under fire. So 20W is max power in my current scenario (not theoretical max power consumption of the CPU in the device).
Once I find time for SQM or other typical router scenarios that show more max power, I would then post an update of that single line or add a second line.

account4538 · September 15, 2022, 3:12am

Everyone keep in mind the power supply itself is a big factor in power consumption. For x86 we have the 80 Plus efficiency ratings: https://www.velocitymicro.com/blog/what-is-psu-efficiency-and-why-is-it-important/ So obviously 2 identical systems will have different power draws.

For embedded and SBC the power supplies are usually the wall worts or external power supplies with the latest required efficiency rating of VI : https://megaelectronics.com/the-difference-between-efficiency-level-vi-and-v/

The efficiency rating also changes depending on the draw of the total capacity of the power supply with around 50% being the sweet spot.

moeller0 · September 15, 2022, 7:29am

Intersting comment. I guess for the actual measurements that is orthogonal as we measure the energy/power before the power supply units and hence the efficiency of the power supplies factor in. One more argument for calculating long term average power numbers, as these will sort of account for non-linearity in PSU efficiency, assuming the sample period contains typical usage patterns, no?

It is tempting to add another column for model and make of the PSU, but I think this is taking things too far... as the "table" is already quite unwieldy.

Speaking of unwieldy, @tmomas had a great suggestion, namely putting the data into a linked editable spreadsheet so that data entry and formatting is less of a chore and would make adding columns a piece of cake.
Anybody opposed (or for) this approach, please let us know.

Pico · September 15, 2022, 1:58pm

for custom x86 builds you would mention a long list of components anyway, one component being the chosen PSU.

for off-the-shelf router/device (x86 or any other) you would just mention nothing about a PSU, then people would get the idea that you have been using the default PSU which was packaged with the device. Even though vendors change such default PSUs frequently, I would assume that there is no noteworthy difference to insist on mentioning it in the list.

if you had purposely chosen to use a custom PSU for an off-the-shelf device, you could just mention the PSU in the details column. People would get the idea to look there if the same device appears several times with different power readings.

But since this has world wide readers, mentioning whether it was US or non US (115V or 230V power outlet) might make a small difference to avoid people scratching their heads about different power readings, as 115V usually has a bit less PSU efficiency.

moeller0 · September 15, 2022, 2:49pm

OK, added that field (voltage @ frequency) and tried to fill in the values. @Arie I filled in 230@50 for you, if this is wrong, please let me know what would be correct. Thanks

dlakelan · September 16, 2022, 5:08pm

Don't have the time to do a ton of data collection, but since it came up in another thread. I have a Zotac Zbox CI 327 running a celeron 3450 which is a bit older, but basically a decent base for an x86 router (it has dual ethernet built in for example, it's fanless, it was pretty cheap). I'm running it as a media pc. At idle my kill-a-watt meter shows between 4 and 5 watts (it kind of oscillates back and forth). Running a speed test it shows 7-9 watts, with a few peaks to about 11 watts. This is actually running firefox browser on the box and doing a waveform bufferbloat test on the box.

Power consumption is the box + its power supply + several USB hubs and various USB devices, but doesn't include the monitor. Also it's power factor corrected, not a VA measurement.

If I were going to guess at its power draw in real life as a headless router, I'd guess less than 5.5 watts long term avg on this basis, and very likely something like 4.5 to 5.

X0rCode · September 17, 2022, 12:25pm

I use:

Draytek 167 Modem
Raspberry Pi 2 ( Adguard )
Belkin RT3200 with openwrt both radios are on.

My total Power consumption is mostly 10-11 Watt all three devices. My devices turn off automatically everyday for 8 hours.

juliank · September 17, 2022, 2:49pm

For comparison sake, quad core ARM Cortex-A53 processor that is in many AX routers is more than 2x faster then Celeron 3450 in single and multicore benchmarks.