An old PC draws easily 5-10 times more energy.
Depending on your location, the yearly cost of running a Pi is around ~10$. The big machine then 50-100$. So energy wise, a small power efficient machine might be more expensive but the running cost could be lower.
This is really small fry compared to other HVAC efficiency concerns, and definitely not an issue outside of summertime temps in most locales.
My Threadripper on the other hand - I had to move that into the crawlspace as it was: a) loud as hell, and b) basically a space heater that also does useful compute. 160w idle, ~280w full tilt - that thing is very noticable.
Just as a reference point, straight up "electric heat" is basically 250 watts A Foot.
It makes a big difference in the summer, but I miss it during the winter.
BTW, I'm using a M.2 2280 drive, even though the hat is designed for shorter drives. (I just tied mine down.) I installed it over the CPU fan and it works great.
I have a full on Dell Optiplex 3070 (i5-9500, 1x16GB memory, 512GB NVMe) running windows 10 that idles at 8W.
I have a lenovo m92p tiny (i5-3570s, released 2012) that idles at 6W.
It has an i5-6500, 32 GB RAM, 2 SATA SSDs and a 4-port i350 NIC (all ports up). Idle means OpnSense and HomeAssistant running inside KVM on top of whatever kernel version was current in Arch at the time, but with no traffic.
Does the raspberry pi draw 1-3W only? It should be noted that old pcs like these can be had extremely cheap, so the difference in price should take this into account. Moreover, if you need extensions of any kind (NICs, drives), getting them running at all on a PI is somewhat more involved than on a standard PC.
In either case, USFF is an order of magnitude less energy than desktop so it's still a win most of the time.
What do you mean by this?
> In either case, USFF is an order of magnitude less energy than desktop so it's still a win most of the time.
Doesn't it depend on the actual CPU used? I have an HP Elitedesk Mini (which is basically a small laptop without a screen — the equivalent of the ThinkCentres mentioned in this thread) and the CPU is an i5-8500 IIRC. I don't think this particular configuration would draw much less power than a "regular" desktop / SFF, aside maybe from the RAM (it uses SO-DIMM). I've never bothered to measure its idle power draw, and don't have any comparable "full desktop" to compare.
I don't know if HP or Lenovo have models with laptop-class CPUs in this form factor, but I image that would be an actual improvement on the power draw. I did see however Chinese brands on Amazon sell models with those kinds of CPUs.
You're correct that some of these machines have desktop CPUs, what I discovered was they also have different classes of desktop CPUs.
The m920q/m920s appear to have differnet power usages an possibly different cpus, or at least they're configured different.
If I compare the two:
- m920q (USFF) can go as high as 135w: https://psref.lenovo.com/syspool/Sys/PDF/ThinkCentre/ThinkCe...
- m920s SFF can go as high as 260w: https://psref.lenovo.com/syspool/Sys/PDF/ThinkCentre/ThinkCe...
Additionally, if you look up the Gen 6/7/8 power usage for the cpu itself, you can see the idle and total is different as well.
The trick is Lenovo is far from the best priced USFF in this category for the same thing.
When I bought Lenovo, I wanted to try out Proxmox, had no interest in fighting with installation or drivers, an just picked something that had flawless installation... that has since improved with other manufacturers.
The tool from OP here is really cool - I'd probably add some from the spreadsheet I had built out but I haven't had to buy many more of these once I figured it out. Load them up with ram, a few ssds, a UPS and they're pretty solid.
Dell Wyse 5070 with Pentium Silver J5005 ~ 5W
Fujitsu Futro S940 with J5005 as well ~ 7W
Dell Optiplex 3080 Micro with i5-10500t ~ 12W with two SSD's
In comparison, my Ryzen 7 server build consumes about 22W idle (before I added GPU), has 4x SSD and 4x RAM sticks. I like raspberry pi, but for most purposes an used mini-pc is a better choice.
Obviously, Raspberry Pis, SFF boxes, workstations and rack mounted servers all occupy different niches (with some overlap). Anyone confidently stating that one could fully replace anther with no context of the workloads is wrong.
Speaking of my projects - the RPi is perfectly capable of working as a web crawler (at a page rate that may surprise you) as well as a media download client & transcoder (again, simultaneously transcoding a number of streams that may surprise you).
My homelab includes 1L PCs and 1 U rackmounts (actual servers) - I appreciate what each brings to the table.
The reason I prefer mini-pc's over pi is the x86 architecture and possibility to add more RAM. For maximum flexibility, I mostly run my self-hosted services inside virtual machines that I manage with Proxmox, and pi isn't ideal for that. Admittedly I found even the mini-pc's too limited due to lack of space/pcie slots for a GPU, and ended up with a custom desktop build. That allows me to experiment with stuff like self-hosted AI, and game remotely. Support for ECC RAM and more SSD's was a big plus too.
So, indeed it all depends on what and how much you want to do with the machines.
It could probably run a single-task relatively well, like PiHole or something, but otherwise it's in a completely different performance category. Like an order of magnitude.
So 6W idle for J5005 would put it on the same level of efficiency.
I recently (8 months ago) replaced my 10 year-old laptop. The only reason I retired it was because the display was starting to go.
So I bought a second-hand workstation-class laptop with 6 beefy CPU cores and kinda wish I hadn't. Overall I want to like it but the battery life is abysmal, it makes a lot of heat even when fairly idle, and is a bit heavy due to the large heatsink inside. (And that's without a dedicated GPU.)
If I had to do it over again, I would trade it for one with a weaker but more power-efficient CPU.