99.997% of a routers time is idle. The J4125 processor of the machine I linked above has a thermal design power of 10W, so if you run that whole box computing pi to a billion digits on max tight loop the whole box would probably use 18W including inefficiency in the power supply. In reality, it'll be idle as a router, and the whole thing will probably use 3-5W out of the box.
1 Like
In my experience, many OpenWRT users on dedicated devices do so precisely to run application stacks atop them, well beyond the realm of simple routing, so I'm not sure that's an apt comparison.
Regardless, my intent was merely to answer the OP's question about reasons people choose the gear they do, and my position is that people buy 'overpriced routers' for the same reason they chose automatic transmissions and LED bulbs when they were the more expensive choice...simplicity and indirect savings.
We can debate one-off hybrid use cases all day long, and all it's ultimately going to prove is what we already knew...everyone's needs and motivations are somewhat (or hugely) different.
2 Likes
You're not wrong. But I do think people tend to be unaware of the relative ease of use and availability of low powered mini-pc x86 devices. Single small package low powered (5-10 watt) x86 devices are a relatively new thing, maybe since 2016 or something.
3 Likes
Speaking of which, and sorry for going on a tangent, I occasionally keep thinking of those "PCs on a stick". Inherent lack of ethernet ports aside, I wonder if they would make a worthwhile target as very compact and quite powerful (travel) routers.
1 Like
WLAN is probably their weak point, Realtek sdio was popular in that sector.
How that? If the energy demands of the base functionality already differs between a consolidated all in one and a decentralized BYOD solution, how is adding essentially the same number(s) to both sides of the inequality change that absolute difference?
I grant you that the relative difference between the solutions will get smaller, but I for one have to pay my absolute energy consumption, so X awatts more or less makes a noticeable difference to me.
Full disclosure, I am running a mixture of consolidated and decentralized:
bt homehub5a as bridged modem ~7W
turris omnia as dualband WiFi router ~8 W (running SQM/pakon IDS)
gigaset c610ipa, VoIP basestation ~1.4W
total: ~16-17 W
This already consumes more than a modern top of the line all-in-one modem-WiFi-router like a Fritzbox 7590 (estimated ~11-15W depending on usage patterns). I doubt that exchanging the omnia with a wired only router an a dual band AP would not add noticeably more to this difference...
(These are averaged numbers over multiple hours to days, so reflect the average consumption well. I prefer longer term averages over simply trusting idle power values taken from datasheets, but then I can measure longer term energy consumption reasonably well, instantaneous power less so).
Sidenote: Over here is an attempt to collect typical energy consumption numbers for devices in a home network, to allow better informed decisions when taking expected energy consumption into accpunt when designing/extending a home network.
I think there is one more thing worth mentioning here - extra devices you need with x86 to have the same functionality integrated routers have.
At the very least you need a switch, and very likely - AP, because good wireless cards suitable for using as AP are (almost) non-existent.
So we have one small box vs at least 3 small boxes, with both AP and switch consuming power too, in fact very similar to power consumption of mid-range integrated router.
x86 (or general purpose arm, like pi4) setup is always going to be more involved, larger and consume more power. And IMO it makes a lot of sense to choose based on actual needs, not on what's theoretically "better". If single integrated router is sufficient for a given use case, in terms of performance, wifi and switch ports - the why choose more complicated and expensive setup? On the other hand once certain threshold is exceeded it starts to make sense because "high-end" integrated routers are indeed overpriced for what they are...
Nowadays this threshold seems to be somewhere around ~400-500Mb, which is sufficient for most home users. Yes, providers can offer 1Gb or more, but it is important to remember that this bandwidths is non-guaranteed/shared and having "1Gb" does not mean router capable of 1Gb is strictly required.
1 Like
Citation needed ;). I agree that the majority of time will be spent idling, but this are a lot of 9s.
TDP is a measurement to select/pick the cooling solution for a CPU, it does neither reflect the max. power draw of a CPU (albeit longterm the power draw should converge on this) nor the average power draw of the full system. Intel and AMB both started slightly re-defining these numbers some years ago, so these are even less easy to interpret. That said, they are still somewhat predictive, however with quite a margin of uncertainty.
I would love to see longer term wall plug energy/average power measurements. Not because I do not believe you, but because I think it would be nice to document this somewhere. In case you have measured data maybe post it over in the "energy/power" thread:
However celerons are not necessarily based on x86 big cores and can underwhelm depending on the load (however that can be partly overcome by having enough cores running at high enough frequency).
Usually it is defined as W/K also…
Basically if the cooling system dissipates less heat than 10W for each degree rise in Kelvin the CPU will burn up.
1 Like
But the total cpu power has gone down over time.
My latest “big” computer with Intel i7 11 gen cpu with built in GPU doesn’t even start the fans any more for light office work.
1 Like
Sure, smaller processes still offer more efficient processing (or faster clocks) CPU makers thankfully have started paying more attention to power consumption. However x86 cores can still consume quite a lot when asked, which IMHO makes it relevant to actually measure power consumption under one's typical usage scenarios to get a decent estimate of power/energy and related costs.
1 Like
I was hoping not having to do a detailed analysis and keeping it rather generic, as it gets complicated quickly… 
On a high level overview, the more additional networking infrastructure you need (anyways), the more the power consumption of a single piece (the router itself) drowns in the overall tally - but that's not my main argument here.
Let's postulate a few basic conditions from the start:
- a Fritz!Box like all-in-one device will always win in terms of power consumption, yes it will chug 15-20 watts, but if it really remains the only networking device in use, it sticks to that usage only.
- I consider 'modem' and 'telephony' external to the power usage, as -with OpenWrt in mind- neither will be served by the OpenWrt router, so in the electricity tally it doesn't make a difference if the same modem is used in front of an OpenWrt x86_64 router or a plastic router
- yes, I'm obviously aware of the bthub5, but as you've noticed yourself, it's very borderline as an all-in-one modem-wireless-router
- yes, I'm aware of chan_lantiq, but asterisk "is not for everyone" (to put it mildly), so another solution is probably wanted anyways
- --> there isn't really a lantiq vr9 device like a bthub5 with FXS ports
- the bthub5 is relatively unique in providing good wifi (ath9k/ ath10k), most lantiq all-in-one device cheap out on this topic and go with ralink or lantiq WAVE300 (neither of which are usable)
- FXS ports mean one core gets reserved for the voice core, in other words you lose half of the performance - which is already lacking in SMP configuration
- AVM's devices aren't supported by chan_lantiq
- the 7490 isn't supported by OpenWrt (yes, I know, external development that gets it working - not merged yet, won't get FXS/ DECT working either)
- the easybox 904xDSL is its own world of pain (not officially supported, very special hardware in many different tangents)
- this is considering x86_64 for "fast" connections, let's say >=300 MBit/s (cable/ fibre), which are hard to get working on OpenWrt with 'standard' plastic routers - x86_64 obviously does not make sense for a 100/40 MBit/s VDSL contract, which can easily be taken care of by ath79, mt7621a and just about anything else.
- this also means the modem is external to the electricity tally, as ONT/ cable modem can't be done by an OpenWrt device anyways
- to quote myself, I'm considering the case of "and once you cross the border of 4+1 ethernet ports and a single AP" here, so the need of an external switch (at least one) and an additional AP (at least one)
- neither if which is an 'uncommon' scenario' for an enthusiast household, even a small(er) one
- as we are looking at faster WAN connections (>300 MBit/s), I am comparing against modern/ faster wireless routers - not an old mips getting along with 5 watts
- ipq806x (cortex a15) uses around 15 watts idle, that's kind of the ballpark I am basing this calculation on
- mvebu (cortex a9) might be better, but I'm disqualifying it here because of mwlwifi (yes, the Turris Omnia is special, but also quite expensive)
- ipq807x can be better, depending on its wifi/ wired specs
- I don't have values for mt7622bv+mt7915, but I'd guess >=10 watts
- I assume 5-11 watts for an x86_64 router here, which is a reasonable figure - a modern 4-port/ Atom based one should be close to the lower end of that range, a cheap/ used device closer to the upper end of it
- there is very little difference between a 5-port 1 GBit/s switch and an 8-port 1 GBit/s switch, nor is there a significant difference between a managed- and an unmanaged switch of the same port count, in terms of its power consumption - you do have to skip multiple sizes, to get a noticable delta
- the number of active ports matter more, than the number of ports in total, assume 1.0-1.3 watts per active (link-up) 1000BASE-T port, less for 100BASE-T, considerably more with 5GBASE-T or 10GBASE-T
- EEE helps a bit, if supported on both ends
- some devices drop to 100BASE-T on link-down
- in terms of the overall tally, it doesn't matter much if a port is on your router or your external switch
- a modern (post 2015) 1 GBit/s switch will be much more energy efficient than an older one (pre 2010, often with transformer based PSU, no EEE, early silicon)
- the number of active ports matter more, than the number of ports in total, assume 1.0-1.3 watts per active (link-up) 1000BASE-T port, less for 100BASE-T, considerably more with 5GBASE-T or 10GBASE-T
So let's look at the hardware that can do >300 MBit/s (maybe with SQM):
- x86_64
- RPi4
- NanoPi r4s/ r5s
- with sqm maybe up to 500 MBit/s, give or take
- mvebu
- mostly disqualified because of mwlwifi (yes, I'm aware of the specific QCA wifi in the Turris Omnia, but the Linksys WRT range is by far the more common device)
- I'm guessing about the sqm performance here, early benchmarks suggested 1 GBit/s, but DSA has dropped full-duplex performance and modern -non-artificial-benchmark- uses might be more demanding, what about ~600 MBit/s?
- mt7622bv
- with sqm maybe up to 500 MBit/s, give or take
- ipq806x, I wouldn't recommend it above 500 MBit/s without sqm (maybe up to 650 under ideal circumstances), not more than 180-190 MBit/s with sqm (PPPoE might drop the numbers further)
- ipq807x, this is still hampered by the 'unoptimized' switch drivers (in the absence of NSS offloading), about 600 MBit/s without sqm, I didn't measure sqm on this platform (as the main slow-down comes from the switch drivers, and the four cortex a53 cores being quite fast, there might not be than much of a dip here)
- filogic 830 isn't really here, yet (should be a good speedup relative to mt8722bv+mt7915, bringing the specs in line with ipq807x)
So roughly speaking, below 500 MBit/s you may get away with one of these plastic wireless routers (mt7622bv+mt7915 or ipq807x), above that you're looking at x86_64, RPi4(, maybe NanoPi r4s/ r5s).
Now, what is the power consumption of a RPi4 with PSU, USB3 1000BASE-T NIC and PSU?
--> my guess is around 4-5 watts plus, ergo not that much different from the 6-7 watts of modern Atom x86_64 - everything else being the same (external switch needed, external AP needed); this figures for a NanoPi r4s/ r5s or dfrobot base board plus RPi CM4 shouldn't be much different either.
If a mt7622bv/ filogic 830/ ipq807x wireless router gets you by, it might win - but not by that much.
So assuming you have a fast line and 3+ discrete networking devices (router, switch, AP) plus modem/ phone, does 6-10 watts for a x86_64 device (compared to something like a RPi4/ r4s/ r5s) really make a tangible difference in terms of median electricity needs?
Details do heavily depend on your your exact environment and requirements, for sub 200 MBit/s WAN speeds you may get by with a single plastic router - beyond that, the air is getting thinner.
For a new 6-10 watts Atom device with four 1000BASE-T you're looking at ~250 EUR from Bezos' Inc. or Jack Ma's market place, a used device like Tips for getting cheap used x86-based firewall with full Gbit NAT (a PC Engines APU) if you are in the US (sophos, gateproct, cyberoam, barracuda, etc. - baytrail-d Atom based, four 1000BASE-T ports, 30g/4096) goes for 30-100 EUR, depending on your patience and persistence.
2 Likes
Certainly it seems very much a sign of the times that power consumption of even networking devices seems to be a significant consideration.
Thanks in part to various green initiatives now all of our energy prices have spiked. Thankfully in the UK we are not so dependent on Russian gas though. Did I read correctly that Germany is now going nuclear?
Well, there is xrx200 (getting a bit long in the tooth), but there is also VRX518 coming along with a somewhat more modern quad-core arm a9 CPU. So modem is debatable IMHO. For VoIP I tend to agree, getting asterisk up and running does seem like quite a chore, especially given that my bast station only takes an average 1.4 W...
But sure the fewer functionality is handled by an all-in-one (or maybe better a some-in-one?) and the more separate devices are in use the smaller the advantage of the combined device gets....
Not living in the 80s anymore my "users" expect wireless dect phones for better or worse, so FXS ports do not matter (for my network, obviously not speaking for other or even all OpenWrt users).
Always wondered how to look at the two processing components of the MIPS CPU, full autonomous cores or more like SMT siblings?
Although single core ath79 will not allow traffic shaping close to 100/40 combined... Been there tried that (the BTHH5a was even worse, due to the CPU demands on operating the modem, nice device but IMHO already at its limit as all-in-one at around 50/10 Mbps).
Well, this brings up the IMHO interesting question whether/how much energy can be saved by using an SFP-ONT in a router with an SFP cage (like e.g. a turris omnia)?
I accept your point, but raise you the turros omnia with 5 LAN ports plus 1 WAN/SFP port ;(
But again a typical AIO with its 4 LAN ports might save you from going for a 16 port switch or 2 8 port ones...
Mine currently clocks in with ~8Watt average, but it is not working that hard on a 105/36 shaper limited link. This is with using both radios but not not excessively (I took pains to measure normal usage and neither artificially quit or busy network).
I would love to see longer term average power numbers for different ATOM SoCs, not becsuse I doubt you, but because I believe that there are still differences and not all Atoms are equally suitable. Not that I have data to support that hunch.
Yes, but there is a difference between an AIO'r 4 ports plus an 8 port switch versus an x86 with two 8 port switches or a 16 port switch. The question is, how many ports will be needed above 4.
Personally I wish it would work better, I have seen issues that required disabling EEE in the past. I ike the sentimet, but the reality so far has been underwhelming.
Bidirectional traffic shaping on an omnia topped out at 550/550 when I tested a few years ago, the 1 Gbps numbers floating around are at best unidirectional tests. I would guess 300/300 is well within my omnia's wheelhouse, but 500/500 probably only after reducing the number of services I run/use on the router.
But that is not really the point here. If you go into the >= 500 Mbps WAN territory I think we all agree that AIO is not a real option anymore due to lack of affordable options. However the initial claim read like "anybody still using a plastic AIO" should question the rationality of doing so", and I think that for low enough WAN speeds that can still be a rationally defensible decision.
Probably, yet I would love to actually see numbers here. (Though I would guess 4-5 watts for a pi routing @1Gbps might be too stingy a guess). Again there is no question that at the wan speeds we are talking about now exusting cheap AIOs are out of their league.
+1 however that is a considerably more detailed and data-supported position than the one I interpreted the first post to take.
Yes, with electric current prices doubling or quadrupling saving electricity has a much more immediate return on starting-to-bother-abou-and-replace/retire-overgreedy-devices
I would not necessarily call what is happening in south eastern Europe right now a "green initiative" however it is the driver of the recent massive power price surges.
No, what is dicussed is to extend the period of shutting down fission power in Germany. The discussion however is quite nuanced and contains a number of opposite considerations to bring to agreement (e.g. for this winter having more base-load electricity sounds like a great idea, but then extending the operation of fission plants that had their last mandatory safety check waived a few years ago on the basis of them reaching EOL soon, does open questions about how long one should operate a nuclear power plant without appropriate safety checks).
IMHO fission is only ever a worthwhile enterprise if one needs weapons grade material, and I still hope Germany is not...
1 Like
That's fine, but I left DECT capabilities out of the equation, as there is no DECT subsystem in linux, nor any drivers (fully proprietary - even if all commercial DECT bases are linux based), so there's zero chance to get DECT working on OpenWrt anytime soon (realistically 'ever', even if proof of concept code existed two decades ago) - while analogue FXS is possible, via chan_lantiq/ vr9 and asterisk (I tried it and got it working, but I didn't dare to expose it to the open internet), even ISDN would (have been) possible via mISDN.
Disclaimer: I do need/ use one analogue FXS port for a fax machine, several DECT handsets and a Cisco CP-7962g desk phone with SIP firmware.
1 Like
Slightly exaggerated perhaps but seriously for a Celeron even at full gigabit routing it's probably got one core fully clocked and 4 cores idle. So I still call that basically idle. And 1 full hour a year like that would be .9998 of the time fully idle and .0002 of the time mostly idle.
I do have a kill-a-watt so maybe I'll measure one of my older celerons and shove it into the power thread
Much appreciated! Again, I am not wanting to doubt your sentiment, just that seeing hard data makes it much easier to compare things.
True, but isn't that the case for all X86 routers? I'd roll an external AP anyway.
I just think it would be a nifty target to experiment on.
Yes, but for your specific use case of being a travel router, semi-decent on-device wifi would be useful (as that would be the primary uplink in most cases and also required to feed phone/ notebook as AP concurrently). A weaker CPU is easily offset by good/ reliable wlan (which neither rtl8723bs, nor brcmfmac, nor mwifiex, would be).
Still nice to experiment with.