With power consumption being particularly relevant at the moment, I was just thinking about whether I could streamline my system a bit.
I note that the “best” server for HA is a bit of a contentious issue but I was just looking at whether an NUC or Raspberry Pi 4 might be a better option - I would not be running just Home Assistant on it but also PiHole and TVHeadend.
Purely from a power consumption point of view, how do the Pi4 and NUCs compare?
Given that they are going to be left on all the time, wattage is more relevant than ever.
My research suggests that the gap between the two is actually less than I had first thought but comparisons between the two are really quite difficult to find.
I run my HA on my home PC (passive cooling but with an i5 processor, 16GB RAM, an SSD etc.etc.)
It idles (with HA running) at around 16 Watts.
I am guessing a Pi4 (if you include the power pack) will use around 6-7 Watts.
So the difference of 10 Watts will add up to around 87 kWh in one year - which in my case results in an additional cost of around 15.- dollars.
Money well spent in my opinion
By the way - the TDP of a processor tells you very little about how much your system will use when it is running HA. The only way to find out, is to measure the energy usage of your setup via a multi-meter. a Shelly PM , etc. while executing HA.
If power consumption is a concern using thin clients is a better solution IMO unless you also want to run heavy CPU/GPU intensive addons like Frigate. Mine has only a slightly higher consumption than a pi4 and as a bonus runs cooler and has way more processing power.
I actually had to look up what “thin clients” was and realise that NUCs are just a brand of mini PC, so I should be open to anything similar.
I guess I should do a bit or real world testing with a wattage meter. At the moment, I am running HA on a RPi3B+, PiHole on a Pi1B and then a separate server (Odroid HC2, with a 14TB 3.5" HDD) for media streaming and TVHeadend recording. I would have guessed that all three of these devices would use more energy in total than a single device but perhaps I should actually test what that combined energy consumption is.
Ideally, if I did go down the route of a single server to do everything, I would be looking to get something with 8GB.
I hadn’t realised that Odroid do some alternatives to the RPi4 with 8GB.
The M1 looked pretty good until I realised that its power sata connector only provides 5v (so no 3.5" hard drive).
The Odroid H2+ looked interesting, as it had two full sata ports, but that has since been discontinued, which is a bit of a shame.
What NUC do you have i have an old NUC6i5SYK and thinking of upgrading from my PI4 but do not know what the difference would be in energy and power of my HA in a NUC.
A ideal Raspberry Pi 4 setup would consume around 10watts all of that with a zigbee usb dongle and a SATA SSD connected to a USB adapter.
while a cheap Dell Optiplex 3020M consumes around 10-12watts (average of 11watts power consumption for 24hours), of course the power consumption peak at 30watts specially when working with ESPHome and compiling firmware stuff. But this one nice to have things since you basically have to wait around 1-3 minute to compile vs a rpi4 that takes 15-30 mins just to compile 1 firmware. This is base on baremetal install.
The Pi is plugged into a Sonoff S31.
It also has a ZWave and a separate ZigBee stick attached to it as well as a 128GB SSD (Kingston Technology Company Inc. OTG V2).
The Micro versions that the Likes of Dell, HP & Lenono are pretty good alternative Nucs as just a touch thinner but with a larger base sort of between a Nuc and Mac Mini. Sometimes called USFF (Ultra Small Form Factor)
I have a Optiplex 3050 Micro with an I3-7100T with a 120Gb M.2 NVME & 1GB 2.5" hardrive and basically they are little square laptopless motherboards in a case ready to go.
Idle I get approx 8watts which with the NVME & Sata 2.5" is pretty good and approx £23.26 for the 70Kwh for 24/365.
About the same processor level are the RK3588 boards and as long as you use microsd or emmc they are 1.5watt idle £4.36 13.5kwh 24/365 but things like NVME are not great for idle and not sure why… (or the ones I have)
I like both and prefer the Micro’s to Nucs as they tend to be ex commercial PC’s and for similar spec cheaper than Nucs but a different form factor.
Rockchip did a pretty amazing (Arm really as the newer A76/A55) is near x5 a Pi4 but posting very similar idle/max power draw and can be cooled passive or active and also same with price.
I am running home automation on NUC8i3BEH with 256GB ssd and 8GB DDR4. Which is quite powerful computer - measured from power outlet it consumes 6W when idling in linux. Prior to it I had NUC 7CJYH with 4GB DDR4 and 256GB SSD - measured from outlet 4W. It really depends on RAM installed, each 8GB is 3W more.
Regarding noise and fans. Both NUCs were cooled passively (fans off) for standard linux sensor collecting workload (few dockers are running, about 1GB RAM in use ). In other words mostly idling. Of course if performance is needed fans spins up because this NUC8i3BEH with i3 8109 CPU has enough power to run even WIN11 quite comfortably,.
I would expect that with eg 12th or 13th gen intel cpus you get even bit lower idling power consumption,
My 4GB-RAM Pi4 setup with SONOFF usb dongle (Zigbee) idles at around 3.2W (measured at the wall with a brennenstuhl PM231E). This with an Anker USB-C supply.
No case, no fan, nothing else connected to it, just a Sandisk micro-SD-card.
I see it ‘spin-up’ to 4W here and there. Running HAOS. I’m about to do some days-long measurement to have an average kWh/day.
connected via Ethernet, not WiFi (not sure if WiFi/Bluetooth is on, I guess yes).
I am quite pleased with it’s performance and efficiency. The battery gives it the added perk of an ups and I can detect power outages with it. If you close the laptop’s lid, the screen turns of, simple and effective.
I previously had a Raspberry pi 4 with 4 gigs and a 64 GB microsd card and wanted something with more performance.
My Nuc8i3beh uses an average of 2.9W running Home Assistant. So same consumption as rpi4, but more reliable imho and better response time for more complex cards and dashboards…
