Good point, comrade. App services split to separate list.

Done and done

1. DNS resolver, like pi-hole, unbound with adguard, diversion, etc.
2. RMS server: a lot of Remote Desktop software has the option to install a listener on a low power device elsewhere on the network that can use wake-on-lan to access computers within the network without keeping everything on 24-7.
3. Log aggregator: would be useful for anyone who troubleshoots stuff regularly, but historical info of any kind can come in handy.
   Simplest form might be a scribe server. Network gear often has an option to send logs to a particular URL, so if you added the scribe server IP/port to the field you’d have historical network logs.
4. Additional loggers: could also be run on-device, such as a wifi connectivity checker, smart home or energy monitoring state data, decibel meter with USB microphone
5. RADIUS server for managing enterprise WPA keys
6. Mobile home: due to the size and power draw, when paired with a hotspot and battery the potato could be useful as a mobile service repeater, a VPN client that deploys your home services on the go (e.g. in a vehicle, hotel room, family/friends’ houses, etc) to arbitrary client devices. If you use the same SSID/PW and encryption type, personal devices would use it automatically during travel.
7. Home theater box like kodi or jellyfin client

At the level of individual apps, the list explodes. Many progressive web apps can be hosted essentially for free on the potato, so you could shunt your always-on services to this machine to allow low power states on a beefier machine. For example:

1. Network management or security software like Fing
2. Low throughput NAS or incremental backup management server like rdiff, TimeMachine, etc
3. inventory management like partkeeper, storaji, etc
4. Smart home bridges like homeassistant or homebridge
5. Bookmark aggregator or landing page like heimdall, raindrop, pinalist, etc
6. Retro game emulators or ROM libraries like retropie
7. Photo libraries like photoprism
8. Book libraries like calibre-web

Edit: list subitem formatting messed up
Edit: add common micro services, mobile deployment
Edit: add home theater suggestion
Edit: add always-on and PWA examples

Apple used to straight up steal the idea of existing apps. Lately it seems they favor buyout, like with dark cloud becoming weather, but it used to be that Apple would randomly swoop in and crush developers by creating a first party version of their app.

I think you can also register 10 years in advance, or maybe more depending on the registrar, which would cover all other potential snafus like expired card info.

deleted by creator

Ah! Been there. Allocating lanes on small systems always seems to have more trial and error than I expect.

And here’s that x4 SFP+ card: https://www.trendnet.com/products/10g-sfp-pcie-adapter/10-gigabit-pcie-sfp-network-adapter-TEG-10GECSFP-v2

can’t find any SFP+ card that is x4

Pretty sure mine is x4. I’ll check.

Edit: confirmed.

deleted by creator

+1, also chocolatey

Maybe yeah. Also got the sense from the strong opinions that this is a preexisting debate, presumably in the context of continuous workloads or cached arrays with minimal spindown intervals. In that context it’s true that rotational disks still often win in energy efficiency and robustness (assuming we’re comparing them to consumer SSDs and not the latest enterprise u.2 stuff that’s rated for continuous work).

Not sure what everyone is arguing about here. Clearly SSD is better for intermittent r/w, whereas HDD can be more efficient at continuous r/w (especially in terms of watts/TB)

Just looking at specs should be enough to see that. SSDs can idle in ready state at close to 0 draw (~0.05w) whereas HDD requires continued rotation to remain ready. So consider an extreme case of writing for 1 minute then maintaining ready state for the rest of the day. For that the SSD will be far more efficient, obviously.

I don’t know, but I’d guess the buffered chipset controller has more stability during certain power state transitions.

deleted by creator

deleted by creator

As a human, honestly I too would have thought there was a CLI package for the HuggingFace API.

Edit: there is (now at least) https://huggingface.co/docs/huggingface_hub/main/en/guides/cli

In the end people/companies will pick the proprietary / closed option just because “it’s easier to use” or some other specific thing that will be good on the short term and very bad on the long term.

I agree with most of the above, just wanted to relay an explanation given to me years ago by my then eng director in an argument about this. He said the reasons we tend to use proprietary / closed platforms and deps in business settings is not necessarily because the software was better or easier to work with. Clearly it often isn’t.

It’s because of (1) built-in factoring and infrastructure, (2) built-in domain expertise that would otherwise require hiring or training, and (3) contractual guarantees that can be invoked when things go wrong. All of which attenuate risk and make development timelines and outcomes more predictable.

His line was “OSS is free like a puppy is free.” That is, most businesses aren’t old enough to handle the responsibility, and that’s why we still sometimes use shitty proprietary software.

deleted by creator

Dammit, I came here hoping to see at least one “I have a very special set of skills.” Oh well.

Yeah I’d cut bait, rebuild from latest tapes. But also…

I’d put the corrupted backups in an eye-catching container, like a Lisa Frank backpack or Barbie lunchbox, to put on the wall in my office as a cautionary tale.

Nice, sounds like you narrowed it down.

You can leave turbo boost on and make more subtle adjustments using command line utilities like cpufreq or with GUI-based unraid plugins like this one.

Before spending time fiddling with settings though, you might try using /sys/devices/system/cpu/cpu*/cpufreq/energy_performance_preference to set one of the built-in profiles like balance_power. If you do need to make manual adjustments, I would try lowering max clock speed first.