memes

It's really down to fitting the machine to one's Requirements, present and forecasted ones.

So my home server is just a N100 Mini PC because it's just a TV Media Box on my living room that doubles as home NAS and Torrent server with a dedicated VPN connection, for which an N100 with not especially large or fast memory and a decent-sized SSD, is more than powerful enough since the CPU heavy stuff - video decoding - is done in dedicated silicon inside the N100 so doesn't really run on the CPU cores, whilst the other functionality is mainly bottlenecked by network speeds and my network is just Gigabit Ethernet.

If I expected heavier CPU loads I would have gone with a different CPU (plus associated elements such as motherboard and memory) whilst if I wanted to run the heavier AI stuff (such as image generation) it would've been a Desktop PC with a dedicated Graphics Card with lots of video memory.

As it is, my games PC doubles as Image generation machine and also works fine if I want play with VMs or Databases since that's running Linux and is a lot more powerful in almost every way (curiously, not disk speed since it's a bit old with upgraded parts, so it's still using SATA and does not support M.2 disks on PCIe) than that Mini PC.

A machine on my living room is supposed to be quiet (so, no loud fans, hence low power consumption), so I was hardly going to over-dimension that living room TV Box / Server just to once in a while I could play with heavy stuff in it, given that I already have a different and much more powerful Linux machine at home that I can use for that, hence why I partitioned my needs this way and can have an always ON server that just tops at 20W (though generally it uses less than half that power).

PS: Also keep in mind that merely running a database isn't by itself any kind of heavy load (even for heavy stuff like Oracle, much less mySQL or PostgresSQL), it's what uses it that dictates the load, so even running a DB there is not an issue unless I'm doing tons of massive non-indexed queries against it (or huge dataset indexed ones, since non-indexed ones on huge datasets end up disk bound unless you have insane amounts of memory) or a similar pattern of usage.