Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Mini PC for local LLMs in 2026

terminalbytes.com

Post reply on HN
Print view View original

21–30 of 36 posts

Re: Mini PC for local LLMs in 2026

#21

Post by pjmlp »

Neywiny wrote: post #6
pjmlp wrote: post #3

Currently NVidia's mini PC, or the version licensed to Asus, is one of the few that I can actually buy with Linux pre-installed with a fully OEM supported version. One would expect that by now buying desktop class computers on shops with a Linux experience would be rather common. Geekcom devices that it advertises as Linux ready, are actually sold with Windows pre-installed. I guess they mean WSL ready.

I would guess they mean it's ready for you to install Linux on it

Yeah, ignoring the whole fragmentation that keeps happening on the desktop stack, The Year of Desktop Linux will never happen if only computer nerds get to build such systems, as it has always been.

Instead normies get The Year of Linux kernel deployed with all kinds of consumer devices, and The Year of Linux VMs on retail.

Re: Mini PC for local LLMs in 2026

#25

Post by jmyeet »

There's some mention of Apple silicon here but it's worth expanding upon. Macs have a unified memory architecture. So if you have a Mac with 64GB of memory then the GPU can use all of that. This is potentially quite useful but Apple silicon in general is limited by memory bandwidth. For comparison, a 5090 is 1792GB/s. Here are some examples:

- GMKTek EVO-X2: 120GB/s reads, 212GB/s writes

- NVidia DGX Spark 273GB/s

- Mac Mini M4 120GB/s but only $600+

- Mac Mini w/ M4 Pro 273GB/s ($2199 for 64GB)

- Mac Studio M4 Max 410GB/s ($3500 for 128GB)

- Mac Studio M3 Ultra 819GB/s ($5500 for 96GB)

- Macbook Pro 16" with M5 Pro 64GB 307GB/s ($3300)

- Macbook Pro 16" with M5 Max 128GB 460GB/s ($5399)

Sadly, Apple discontinued the 512GB Mac Studio. Mac Studios are a little long in the tooth now and due for an upgrade this year. I suspect that prices will be a lot higher given the RAM prices but we'll see.

Re: Mini PC for local LLMs in 2026

#26

Post by edot »

znpy wrote: post #2

As somebody that has a vague interest in running local LLMs… they day i decide to burn cash on hardware I might as well go all-in a get either a 128gb mac studio or an nvidia dgx spark (or some other equivalent gb10-based system). The 64gb mac mini is also interesting, if anything because it is very likely to hold most of its value when reselling. I’m keeping an eye on the next apple hardware refreshes, particularly…

I am in a similar boat to you, but I can’t make the money math work. Local LLMs obviously have a privacy benefit but DeepSeek V4 Flash (which you’ll struggle to get running on any single Mac - you’d need at least 128gb RAM) is $0.14$/mtok input $0.28/mtok output on the API. You’d have to be just absolutely burning tokens to ever make this make sense.

Mac Studio M4 Max with 128gb at $3,699 (if you can find it) would equate to 10 million tokens a day of mixed input-output for over 5 years to break even. At which point that hardware is outdated compared to the SOTA models that will probably still be cheap on hosted platforms.

Re: Mini PC for local LLMs in 2026

#27

Post by amelius »

adityamwagh wrote: post #15
amelius wrote: post #4

Earlier quoted context omitted.

The models are good enough now, so I'm waiting for the day they start selling inference ASICs with 100x the token output speed. See Taalas demo.

Taalas is a nice concept, but I don’t want to use the same model forever!

Just buy a new one every few years, just like your phone and laptop. And sell the old one.

Re: Mini PC for local LLMs in 2026

#29

Post by pjmlp »

mark_l_watson wrote: post #19

I bought a 32G MacMini over two years ago and it has been great for experimenting with local models, and now is even useful for local coding (at a slow speed!) with models supporting large context sizes. With the current extreme RAM shortage I deeply regret not buying a 64G MacMini a few months ago. I bet a zillion people feel the same way.

Which is why the Mac Pro was actually relevant.

Those of us on PC land can at least extend them, or exchange the GPU, even if pricey.

Apple has lost the server and workstation market by their own decisions.

Re: Mini PC for local LLMs in 2026

#30

Post by spwa4 »

bachmeier wrote: post #10

"Local inference is rarely cheaper if you’re being honest with yourself about how much you actually use it." Sorry, but this is not even close to "being honest", it's bad math. That calculation assumes you do nothing with the computer other than local inference.

Huh, you make me curious. Let's actually do that calculation. Let's say you do actually do 24/7/365 AI use. Let's say by some miracle you can do 60 t/s on Qwen 3.6 27b, and let's say this PC cost $3000 (you should be able to do this on a DGX spark, and one of the non-Nvidia models, e.g. the Dell one. $3000 would be a good price, but not totally out of the question). And, of course, let's say these prices remain stable.

So that gets you 1_892_160_000 tokens per year at full blast.

If you go the openrouter, eh, route, you'd get charged $2 per million tokens (anywhere from $2 to $3.6 per million tokens). So the value you'd get from your machine at 100% utilization is 1892 * $2 = $3784 up to 1892 * $3.6 = $6800)

So yeah, not counting electricity and your time the machine "is worth it".

[1] https://openrouter.ai/qwen/qwen3.6-27b/providers

Post reply on HN