If you doing something that is hammering a CPU at 100% 24/7, then you’d get a faster CPI to get the workload completed faster and you’d be back to having idle CPU cycles…
Now RAM on the other hand!
Free RAM is wasted RAM!
Not all workloads can be "completed" by throwing hardware at them. The sort of thing this kind of hardware is ideal for is things like home automation.
I run Home Assistant on an Xeon ITX PC with a micro-PSU for example, it's at ~40-60% CPU and ~80% RAM (of 16GB) running inference on my camera streams, handling Zigbee and Z-Wave devices etc, running automations, handling all of the sensors for the doors, lights etc in my home. This isn't a bursty problem you can simply get a more powerful CPU for, the goal here is to keep power down and performance up because it's a constant load. A smaller, lower power micro-PC would save me a bit in energy costs if I wanted to.
For real work, people like me (us?) use real servers like you say; I run a Ryzen 9 12c/24t, AsRock Rack, 2 M.2 NVMe, HBA, 8 SAS bays, 128GB RAM, 6 GbE NICs, with dozens of VMs and containers on, a few Kubernetes clusters, and a bunch of services (git, Harbor, Argo, etc) it uses about 100-150W with my CPU downspecced to 65W TDP.
Now that's a system that has bursty workloads, but it's in a different class to these low power machines that "normal" home users are looking for.
> Now RAM on the other hand! > Free RAM is wasted RAM!
Can't argue with that.
ZFS will take care of that for you.
And likely isn't correct for the minipc (1st item) either, though I've not got one of those (yet) to measure it.
Even my i3 NUC isn't that much heavier, consuming 15ish W (IIRC again) idle.