I don’t think its rpi or network switch, unless you’ve overclocked rpi with liquid nitrogen 😅. So, I assume its TrueNas device.

If it were a significant power difference, say 20-30 watts, you could easily find the process using htop/iotop. However, 6 watt difference is a relatively small value for a device with ~25 watts of idle power . It might be a process using just 1% system resources. That’s why I would look for systemd timers, cronjobs etc. to find scheduled tasks on specific times. Another possibility is automated S.M.A.R.T. self-tests. Those tests don’t show up in htop or iotop.

UPS devices normally uses wall (input) power, and switches to battery when input voltage is out of the target thresholds. So, input.load should represent the percentage of current wall power (in VA) relative to UPS’s max rated input power (VA). If your devices uses more power, input power from wall should increase as well.

If it’s peaking in certain times, it could be due some scheduled job temporarily increase CPU frequency, or automated tasks like file system snapshot might power-up/spin drives longer than regular usage.

• 2x18k - mirrored ZFS pool.
• 1x47k - 2.5" drive from an old laptop used for torrents, temp data, non-critical pod volumes, application logs etc.
• 1x32k - automated backups from ZFS pool. It’s kinda partial mirror of the main pool.
• 1x18k - (NVME) OS drive, cache volumes for pods.

Instead of single pool, I simply split my drives into tiers: cache, storage, and trash due to limited drive counts. Most R/W goes to the cheap trash and cache disks instead of relatively new and expensive NAS drives.

Biggest difference is being able to execute INSTCMD commands, at least that was the main reason why I developed my own tool. Another less important differences are: older ARM support and since it’s written in Rust, it’s much more efficient in terms of resource usage. TBH, being that efficient only makes sense for very low-power devices.

Besides that, I don’t think you can go wrong with either project.

I’m actively using ollama with docker to run llama2:13b model. It’s generally works fine but heavy on resources as expected.

If I’m recall correct, Sugo 13 only has single 3.5 slot. I’d say look for Sugo 14 (2x3.5) or Node 304 (6x3.5)

Make sure all of your components fits to the planned case including power supply.