I love your detailed explanation of things that I had little to no knowledge of, but I’m also stuck wondering if I’m using my setup correctly, as I have 16 GB of VRAM available but also 64 GB of regular RAM in my computer. Every time I use some setup or some configuration that overlaps or overflows between the two RAMs, my token per second drops significantly, me thinking that it is not set up correctly, so I’m stuck with 9 billion parameter models myself, even though I was under the impression with 64 GB of memory that I would be able to run something like a 30 billion model.

I do run the gpt-oss-20B fast on my 16GB vram, but it’s, limited.

Oh you should be running flash next with strata. Just go to stratas github page [0] and it will explain everything. It runs quite fast on 16+64, you should be able to get 50 tok/s or so. It's a 180b model and it feels a lot smarter than most of the other open weight options. As far as knowing "what will fit" in general that's hard to know these days. The moe models are hard to predict and run wildly different in different engines (llama.cpp, vllm, strata likes, etc) but they tend to be the only good options around 16gb vram. Dense models like qwen3.8 27b are only good if they're entirely in vram. Moe models you need the active count + a few experts (although it will run faster the more experts you can fit).

[0] https://github.com/Niko1221/Strata

I’ll be silly, but having a unique address per device, is indeed a simpler vector to track, right?

On topic, I’m not sure I even trust my own password sanitation, and seeing service A and B hacked every other month, I feel weary. Knowing that someday my passwords might be compromised as well, so what should I do as just a person who can’t remember things? Write them down, I guess.

It's common for devices to choose a new IPv6 address within your range at random to perform a new connection / rotate periodically.

Which in turn makes it harder to track vs. your /64 which is approximates your old NAT'd public IPv4.

It’s combined with UniFi I assume, what else has good API, Hooks, or consumer level integration to play around with with.

Frigate is an excellent open alternative.

I saw a comment and didn’t prompt an LLM and see the whole difference: https://www.diffchecker.com/QD8AKShC/

Same, and hope I can afford that w/$ $/t, 650w is wild to run here24/7 , would cost me 32% of my salary

It typically runs 120W idling, and 450W peak. Not great power efficiency wise, but I have a Cron midnight shutdown and and S5 bios event powering it up on a morning (I could use wake on Lan but didn't see the point).

(650/1000)(24*30)=468 kWh.

That would cost me about $70/month ($0.15/kWh) or someone in California about $234/month ($0.50/kWh).

Do you pay $1/kWh and make ~$1500/month or something? I can’t make the math work for your case.

Exactly, small edge models is the future, highly personal experiences, and not these massive models that the cloud providers currently shove down our throats. While massive models are useful, these massive platforms are about to burst out of their promises. All while we’re supper happy with tiny 4b up to 12b models working amazing for all these “omg ai thinks” daily tasks.

I read this as a phone contract having a bundled: unlimited whatsapp, Spotify, Facebook bundles where data usage is not counted towards your total data allowance in the contract. For example in my country in the middle of Europe you have bundles with unlimited traditional communication and 20 50 80GB of mobile data.

When you WhatsApp all day sending videos, gifs, messages, voice messages, pictures, documents..I can consume up to 12GB per month, this costing me an amount of my 20GB data plan. While a other provider I can choose has a 20GB data plan with a unlimited WhatsApp bundle. Sadly they don’t offer eSIM and 5G so it’s a no go for me.