This year I built a NAS . My focus was to optimize the price not the power, so I planned to go with a raspberry 5 or a raxda 5c because of their lower consumption. For what I gathered a RPI 5 and similar draw 3W idle and 12W at full power and a N100 based computer draw 9W at idle and 24W at full power (approximately of course).
But then I looked at the power consumption of the consumer grade HDD disks. 4 disks would add between 10 and 14W at idle and between 16 and 20W in operation, and suddenly the advantage of the arm based computers in power consumption is less striking.
Moreover you can find on AliExpress N100 mini-pc for 120€ with 16gb RAM and 512gb SSD. Aliexpress is risky but it was much less than the RPI5 with 16GB RAM or just a bit more than the raxda 5C 16GB , both without drive, case and power supply. And the raxda 5C would have been also bought in AliExpress so no almost as risky as my N100.
At the end, for cheaper to buy and not too much more expensive in power consumption I went with the mini-pc. I lost the possibility to use extension cards, especially the one that allows to connect up to 5 HDD, but a 4 port USB HDD dock proved sufficient for my needs.
9W at idle for the N100 is similar to what I get with my N100 based PC. Although under windows it's often sitting at 10-12W.
I also see lower power consumption with an 8th Gen intel Core system.
The small cores that the N100 use are size efficient, but not necessarily power efficient. The N100 chip is just not that efficient power-consumption wise.
I have a regular i5-6500 in an HP 800 G2 SFF. It has 32 GB RAM (2x8+16), two Samsung 840 SSDs and a 4-port i350 network card. It Linux as a KVM host with OpnSense and Home Assistant on top.
According to some watt-meter I got off Amazon it idles at 14W with the 4 interfaces UP but next to no traffic. I consider it idling when the CPU usage as reported by the host is under 5%.
Now the watt-meter isn't some top-of-the-line exotic model, just a random Chinese thingy, but it seems to measure close enough to what I expect some other devices to pull.
I can vouch for the power draw in this type of system. I have a Dell OptiPlex 7070 with i5-9500, 32GB and running Proxmox with a Windows VM and a couple smaller instances. I measure 8w idle from my Kill-a-Watt on the system. I was really surprised at how low it is.
The easy path, as someone suggested, is a kill-a-watt that you just plug it into.
The less intrusive path for some definition of less intrusive is a clamp ammeter if you can expose one of the AC wires (you have to clamp around an individual wire, not both hot and neutral). But then you don't need to unplug the system to measure it.
The third overkill option is to have it plugged into a full-time power monitoring and control device, such as a zigbee home automation plug switch. ;)
This can actually be life threatening if done without proper tools and working fast fuses. So just use a kill-a-watt or any other such tool for safety! NB they're maybe not 100% exact but good enough to give an estimate.
Clamp meters are completely safe. The only risk is if you DIY your power cable so you can clamp one lead, but that's not necessary if you just buy an AC line splitter plug. And those often come with the hot looped around so you can get a 10x reading for better fidelity at lower current draw.
But these days I just skip the clamp meter and throw Ikea Inspelning zigbee plugs anywhere I want power measurement.
i wonder if there's a market for pre-made cables with a "loop" in the hot wire, for using clamp ammeters, or a cable where in addition to the choke ferrule, there's another "ferrule" - current transformer - on the hot wire with "test points", where those test points are to a transformer winding around the hot wire. That way you could just put those cables on things you want to know the power consumption of, rather than having to move a kill-a-watt or whatever.
I could make either, but they wouldn't be "certified", as i'd be either replacing plugs or cutting into the wire itself to add a pigtail.
I can recommend those Third Reality outlets as well, I have about a dozen now and they Just Work with my Zigbee dongle (Sonoff ZBDongle-P) and zigbee2mqtt / Home Assistant setup. I use Home Assistant's Prometheus integration to get the data into VictoriaMetrics, which lets me build Grafana dashboards showing the usage of each plug over time.
I'm using smart plugs that have an open source firmware called tasmota. I can scrape the values using prometheus and can build dashboards with historical data.
how did you get a 6500T to idle so low? Is that CPU power or total system power? I have a T part from the next generation, I don't remember the exact model. It's low but it isn't that low.
The processors themselves all seem to be pretty efficient at idle (there's apparently no real reason to get a T processor. It just caps performance). It mostly depends on everything else. The only place I know of to get this kind of info is this thread[0] where people have been building this spreadsheet[1]. Some threads on servethehome.com also report when some pcie cards have broken power management which can destroy idle efficiency.
Which main board are you using? 5W is very impressive. I'd be surprised if it was even possible to achieve with an ATX board without some serious tuning.
I'm curious to know the details of this as well. 5 watts is absurdly low for an ATX system. I've definitely seen 25 watts before. Assuming you went from a 75% efficient PSU to a 90% efficient one, that'd shave off a few more watts but it won't get you that low.
For perspective, I think the typical NIC consumes several watts of power.
That sounds wrong, efficient intel chips should be very comparable to pi's. Measurements for the futro s740 (j4105 CPU) were [0] around 2-4 W idle when properly configured
+1 to the ability to use PCIe cards - a RaidZ2 array of spinning rust will saturate a 1Gb link without breaking a sweat, and with 10Gb SFP+ cards being so cheap it's a nice upgrade (although if you care about power then a newer card like the Intel x710)
If you want something less "risky", ASRock Industrial has mini PCs which are great. I have an AliExpress N150, fully passive which worked just fine but then I saw the shiny of the Arrow Lake-H platform. Ended up getting a NUC BOX-225H for Opnsense. Way overkill, it's great!
The Arrow Lake-H platform can have up to 28 PCIe lanes where as the N150 gets 9. Something to think about if you want dual NIC + plenty of NVMe drives.
Oh, my bad, I don't recall mentioning price in my post. Rather I mentioned a system that has the PCIe lanes to accomidate 2.5G NICs + 4 NVMe drives without compromises.
I have several asrockrack mobos and this is the first time I am hearing of asrockind, very interesting and there are a TON of different models here
Do any of these asrockind machines have a BMC?
I just got a minisforum s100, it's 4c8g n100 and POE which is pretty sweet. If only it had Intel vpro or some sort of BMC it would be perfect. I can attach a pikvm but looks like I need to do surgery on it to interface with the power switch. WakeOnLan has been really unreliable in my tests so far.
But then I looked at the power consumption of the consumer grade HDD disks. 4 disks would add between 10 and 14W at idle and between 16 and 20W in operation, and suddenly the advantage of the arm based computers in power consumption is less striking.
Moreover you can find on AliExpress N100 mini-pc for 120€ with 16gb RAM and 512gb SSD. Aliexpress is risky but it was much less than the RPI5 with 16GB RAM or just a bit more than the raxda 5C 16GB , both without drive, case and power supply. And the raxda 5C would have been also bought in AliExpress so no almost as risky as my N100.
At the end, for cheaper to buy and not too much more expensive in power consumption I went with the mini-pc. I lost the possibility to use extension cards, especially the one that allows to connect up to 5 HDD, but a 4 port USB HDD dock proved sufficient for my needs.