Top
Best
New

Posted by thm 17 hours ago

Apple caught off guard by AI demand for Mac Mini and Mac Studio(www.macrumors.com)
354 points | 394 commentspage 2
srvmshr 2 hours ago|
Slightly tangential question: Does having a spruced up Mac Studio help learning the ropes of model tuning and/or development as good as having a CUDA device?
darvo31 2 hours ago||
Totally saw this coming. My team just grabbed three Mac Studios for local LLMs; the unified memory is a game-changer.
Danox 6 hours ago||
What’s going on isn’t Apple behind in AI model building I thought I read that somewhere on the MacRumors site in the last two years, that Apple is behind its tech peers and Apple might as well just close the doors.

I always thought Apple was in a good position because unlike their peers they didn’t burn billions of dollars trying to build an AI model that has no financial moat around it.

I still think they’re in a good position in comparison to their tech peers and we will know even more when some of the new computers get into the hands of some of the tech reviewers.

I believe the new computer’s will be pretty good hardware wise what I’m interested in, is the Apple software support for connecting several Mac computers together, and some of the other (new?) software that Apple may have written in house to support those who want to run AI software locally that is just as important as the new hardware.

rglover 6 hours ago|
They're making the smartest possible move: let others burn insane amounts of capital and time finding the quirks and once they see a viable lane, execute.

It's old Steve Jobs logic. Works backwards from the customer experience to the technology (they're the only big player I see doing this).

pertymcpert 3 hours ago||
I feel like they failed upwards into the right strategy. I doubt it was deliberate and not because of the failure of their foundation models to decent.
saejox 7 hours ago||
Most of those people will be dissapointed when they experience Q4 variants of those models getting stuck in loops.

I would wait till the ram crisis is over to fetch a future 64gb ram gpu to run Q8 models. Cloud inference until than.

jms703 6 hours ago||
I'm excited to hear the ram crisis will be over. But will it?
etoxin 7 hours ago||
Switching from Q4 to Q8 was a game changer when I upgraded
hnav 3 hours ago||
Apple trying to sit on two chairs. Make bank selling hypeware while keeping distance to not sully their brand too much.
imagetic 11 hours ago||
No they weren’t.
stetrain 11 hours ago||
The part where people starting buying Mac minis just to run OpenClaw was a pretty sudden movement in the market. These computers aren't powerful enough to run big local LLMs but are still being purchased for AI workflows.

A year ago you could get an M4 Mac mini for $399 on sale and now the same one used goes for over $700. The general AI RAM/SSD spike is part of that but there was also a huge demand spike for small, powerful, desktop machines that could be easily configured with these workflow tools.

nullbio 4 hours ago||
Sad that Apple is stooping to the level of spamming the web with bots and fake news to advertise their products. Are there laws against this sort of thing?
ceejayoz 16 hours ago||
Time to bring back the Xserves, I guess.
Xeoncross 16 hours ago||
If I had to pick a product, I'd say an affordable 32GB mac would be the sweet spot for running local models that function well like Qwen 3.8.

It's true, most people don't run models, but being the default platform for running open weights seems like it has plenty of advantages right now. Just like sales benefited from developers defaulting to MacOS for most open source languages like Ruby, Go, Rust, and TypeScript.

mirekrusin 15 hours ago||
32GB is not enough, it's unified/shared memory, you need to have space for usual system and user apps/services.

64GB+ or dedicated 48GB (2x24 on GPUs) is IMHO absolute minimum.

redox99 10 hours ago||
32GB of fast unified memory is enough for Qwen 3.8 27B.

- 16GB for the weights at Q4

- 9GB for the full 256K context at Q8

- 7GB spare for overhead and system.

The problem is that these Macs have 32GB of slow unified memory.

Edit: I'm thinking of a headless Mac mini, if you meant running it on the same machine you're using of course you'll need more memory, but LLMs are best served from a headless server so that's what I'd recommend.

hawk_ 8 hours ago|||
Is this for setup for agentic coding? Why not also run the IDE compiler etc... on the same machine to use those CPU cores as well?
0x457 6 hours ago|||
Keep in mind that if you want MTP it adds a few gigs. If you use sub-agents it turns already slow generation into even slower generation. Won't be doing any compling (so rust, c and probably go are not avaiable) becase those add memory pressure during compiling.

32gb of unified memory is enough enough for system to be used for anything other than LLM generation.

redox99 7 hours ago|||
You can, you just need a beefier PC, and it's more annoying in terms of noise and heat vs throwing something on your server closet. Plus you don't need to worry about other software stealing resources and whatnot.
qeternity 7 hours ago|||
> Edit: I'm thinking of a headless Mac mini, if you meant running it on the same machine you're using of course you'll need more memory, but LLMs are best served from a headless server so that's what I'd recommend.

What? LLMs are best served from a massive PD disaggregated cluster of B300s connected via NVLink.

If you're running LLMs on a Mac Mini, it's because you want to run local, not because it's the best setup.

shagie 10 hours ago|||
> If I had to pick a product, I'd say an affordable 32GB mac would be the sweet spot for running local models that function well like Qwen 3.8.

https://www.canirun.ai (five months ago: https://news.ycombinator.com/item?id=47363754 377 comments)

tristor 15 hours ago||
> If I had to pick a product, I'd say an affordable 32GB mac would be the sweet spot for running local models that function well like Qwen 3.8.

32GB is not enough RAM. I don't even own a device with less than 36GB at this point, and that device I only have because my employer is being cheap. 64GB is a reasonable starting point for running local LLMs + normal tasks. 128GB let's you really run most smaller models like Qwen 27B and 35BA3B with good context. Even Qwen3.8-Flash-Next runs in 128GB with a 4-bit quant.

32GB would be limited to running models like Gemma4 12B and smaller dense Qwen versions like 9B unless you were using very small quants which damages quality of response.

Xeoncross 12 hours ago||
You are mistaken. I'm running Qwen 3.7 28B 4bit (MLX) with a 200k context window and everything total is 32GB RSS.

Is this the best? No. That's why I said the sweet spot. Getting from 16GB macs to 32GB is perhaps possible. Jumping to 64GB or 128GB as the default is simply unreasonable right now.

e28eta 11 hours ago|||
I have a similar machine, and briefly poked at running a local LLM, but got discouraged after a couple days. The quality, responsiveness, and impact on the rest of the system didn’t seem worth it to me.

What sorts of things are you doing with the local LLM? Anything interactive? Should I take another look?

Xeoncross 8 hours ago||
Yes, 15-30 t/sec is pretty slow for local models so I recommend running local LLM tasks overnight where (vs paid plans) there isn't a risk of chewing through your token budget from a rogue loop or sub-agent. Even if it takes hours, you're sleeping anyway so no concern. herdr + pi works great for this but there are lots of harnesses.
Foobar8568 10 hours ago||||
Memory used : 38GB, and I haven't even started a LLM nor podman, I always fight with memory when using LLM on my mac with 48gb.

And I don't remember to have been able to have pushed to 200k context Qwen 3.6. 3.8 is running on my RTX 5090.

redox99 9 hours ago||
Qwen 27B runs very comfortably on a 5090. You need to use Q4 quants and Q8 KV cache. Here's the math

https://news.ycombinator.com/item?id=49514141

tristor 11 hours ago|||
I assume you mean Qwen 3.8-27B? Yes, you can run this in 32GB of RAM, but it's very context limited. With KV cache compression and other techniques, it's better now than in the past, but I'd still want more RAM, personally.

EDIT to add that you need to reserve 8GB for the system if you don't want to cause problems on macOS, which means 32GB RAM = 24GB max for model + context. It takes 18-19GB to load a 4-bit quant of Qwen3.8-27B, so I'd be really surprised if you can actually get a 200k context window. You need to fit within a 24GB WSS (which is generally a more constrained RSS) to get stable performance on 32GB RAM.

jckahn 8 hours ago||
I run Qwen 3.8 27B just fine on my Mac mini M4 24GB. I use Unsloth's Q3 XXS with 128k context. It successfully completes long horizon tasks with OpenCode.
compounding_it 16 hours ago||
More customers is generally a good problem to have in most businesses. Just that the situation is very paradoxical given the supply shortages.
pmontra 10 hours ago|
If those customers are in the market you want to develop. If they are not, money is money but if it comes from the wrong people it might slow you down.
bilsbie 6 hours ago|
I get the impression they want AI marketing points but don’t actually want people to use local AI on their products.

I have no idea why. They could be so successful if they leaned into local AI.

m463 6 hours ago||
I don't think they're ready for local ai. They have memory + memory bandwidth, that's it.

I also don't think they do "technology". For example, containers have been around for a long time, and apple didn't show up. (I know they have some support now). Imagine an apple-native docker/podman doing something like FROM macos:10.12

I was actually surprised when they did their own chips. I figure it was about control.

jonplackett 6 hours ago||
I think they’re too scared to ‘own’ it - it would be someone else’s model and potential security issue.

But apple have to own everything they do so they’re in a bind

More comments...