Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Apple Stops Accepting Orders for Some Mac Mini and Mac Studio Models
- Thread starter MacRumors
- Start date
- Sort by reaction score
I'd love to know if Apple are actually short of memory or are indeed releasing the M5 Studio/Mini soon.
I'm desperate to buy an M5 Studio but would settle short term with a used M4 Macbook Pro Max or a new M5 Macbook Pro Max which I'd sell on if the Studio does come out.
I just don't want to jump the gun and get stuck with an expensive laptop which could be hard to sell for a decent price but at the same time, if the Studio doesn't come out, there could be big delays on even Macbooks with 64GB and over of memory If Apple are struggling to get memory.
I'm desperate to buy an M5 Studio but would settle short term with a used M4 Macbook Pro Max or a new M5 Macbook Pro Max which I'd sell on if the Studio does come out.
I just don't want to jump the gun and get stuck with an expensive laptop which could be hard to sell for a decent price but at the same time, if the Studio doesn't come out, there could be big delays on even Macbooks with 64GB and over of memory If Apple are struggling to get memory.
Steve Job's history of success can be pretty much summed up as "the second mouse gets the cheese". Sometimes, Apple were the first mouse too...
The Apple I & II weren't the first personal computers.
The Mac wasn't the first GUI computer (and the Lisa failed).
The iPod wasn't the first portable digitsl music player.
The market for the iTunes store was proven by "pirate" digital music sites.
The iPhone wasn't the first smartphone.
The Mac Mini was the Mac Cube done right
(post-Jobs example) the Mac Studio is the Trashcan done right.
I think Apple may be playing AI just right by limiting their exposure to the current AI bubble.
AI service providers will need "consumers" to have suitable front-end machines - phones, tablets and laptops - with modest, power efficient local NPU power - Apple Silicon is perfect for that, and I'm guessing the iPhone is well up the list of client devices for AI services. That's good business from Apple. Doesn't really matter for now if the models were built on NVIDIA or if the online backends are being provided by non-Mac "big iron".
If Apple want to offer "branded" AI services, they can license stuff from Google, OpenAI etc. without getting into a spitting contest with them. As they are doing with Siri.
The current AI bubble is widely predicted to burst spectacularly and leave only a few players standing. People will still need phones and laptops snd, hopefully, the AI industry will rebuild itself in a more sustainable and demand-led way. At that point, there will probably be plenty of fire sales of companies, data centres and IP if Apple wants to pick up a bargain.
I don't see where that's going to fit or what the specs would be.
We'll see what the prices & options of M5 Pro mini vs. M5 Max Studio are, and what the Ultra option is, but over the history of the Mini "Pro" a fully tricked-out Mini Pro has usually been poor bangs-per-buck vs. the base Studio.
Currently, by the time you add the best CPU/GPU to the M4 Pro Mini you're at $1600 (for 24GB) vs. $2000 for the M4 Max Studio (36 GB) - which gets you an all-round better processor, better I/O and display support and better thermals. The confusing factor is that you can't configure a 36GB Mini to compare like-for-like, but if you could it would be about $1900 - since Apple's RAM prices are fairly consistent. The only point I see of the higher end Mini Pro configurations is if you want to prioritise RAM over CPU/GPU power.
I ordered that exactly: 24GB/1TB M4. It is minimally six weeks out. I would consider this a ‘common’ configuration, and, as you said, it should suffice for most people and their needs. With the ability to easily stack/add additional SSDs, the rest of the setup should last for years, providing digital rewards that are enjoyable and expedient.
Openclaw has created a niche.
Now people are going to be buying the lower end MBAs and MBPs to fill it, because the Minis are gone.
Now people are going to be buying the lower end MBAs and MBPs to fill it, because the Minis are gone.
Maybe they're killing all of the desktops, not just the Mac Pro. Would be par for the course for a phone company
anthony13
macrumors 65816
I was hoping for a surprise refresh this morning but nothing. I guess we are waiting till June if you listen to the rumors, but it's crazy to me how far away that is when you basically can no longer buy a top tier machine currently.
BSDnostalgia
macrumors 6502
Look at it this way: anything that they sell at Costco has to have wide enough appeal and be making enough money to be worth it. It they ever sold Mac Pro towers at Costco, it must have been a long time ago. The Mini and the Studio fit very easily on that proverbial one table that holds all of Apple's consumer products. Why on Earth would they discontinue desktop products that are all part of the same continuum from iPhone to MB to Studio, and, that all fit on a single large table?
OTOH, the very large increases in the price of RAM and flash will force Apple to do something. My guess is that they are hammering out the details right now regarding how they are going to stage this to try to make this as palatable as possible without just slamming everybody with huge price increases that will shut down sales almost completely.
SpanishAppleNerd
macrumors 6502
I would like to use them as one more tool for my job, yes. So top of the line M5 Pro 64 GB mac Mini it is then?
SpanishAppleNerd
macrumors 6502
Do you happen to have any good recommendations re AI models for academic research?
As much as I want to be in the "local models" crowd, you will almost certainly save more money and get better results just by paying for Claude or ChatGPT. There are probably some use cases where local stuff can work OK (transcription?) and maybe some areas for privacy… but even 128GB is minuscule for LLMs and you will pay a lot for that in your mini (or Studio).
Maybe I'll be wrong as things get more optimized, but the memory companies are bringing new plants online and this dearth won't last forever. In 2-3 years the memory scene might mean you can get a far more capable machine for less than today… so I'm guessing it's not worth buying stacked machines now vs. what will be on the table later.
Maybe I'll be wrong as things get more optimized, but the memory companies are bringing new plants online and this dearth won't last forever. In 2-3 years the memory scene might mean you can get a far more capable machine for less than today… so I'm guessing it's not worth buying stacked machines now vs. what will be on the table later.
IvyKing
macrumors 6502a
Since the current Neo only has 8GB ram and the next version is rumored to have 12GB, I would imagine that Apple would be looking for ways to reduce software RAM requirements.
Thanks. I didn't know there was a change. That bandwidth change is significant.
I could really get into LLMs as a hobby, but I'd want the Max and Ultra class chips for the reasons you cite, and those are just way outside my budget.
Wahlstrm
macrumors 6502a
It’s hard to answer for sure with out knowing what models you want to run and how much else you’d like to have running at the same time.
But if you’re happy running Gemma 4, then 64GB should be plenty for the 26b model but you should probably be looking at closer to 128GB for the best performance with the 31b model.
SpanishAppleNerd
macrumors 6502
I have 24 gigabyte M5 Pro (regular MacBook Pro). The models I have tried are terrible. Either very slow, or usually wrong.
SpanishAppleNerd
macrumors 6502
I am thinking about buying a 48 GB M5 Pro Mini now. 64 is my upper limit, I still think that for my needs either 48 or 64 gigs is more than enough
nateo200
macrumors 68030
Yeah me and my buddies are going back and forth on whether the 64GB's of memory is sufficient for the really insane models one of my friends likes. I think even with only 48GB's you can do some significant work but to really have room to stretch your legs 96GB's would be really nice. For the models my buddy does he needs more memory than he needs raw horse power but unfortunately to get more than 64GB's of memory you need to get a Mac Studio perhaps that will change.
SpanishAppleNerd
macrumors 6502
Not my case, I am an economist so the furthest I go regarding computing power is using SPSS (of course, big data economists go way further but that's not my alley)
A mistake using OpenClaw in that manner shown in the article is quite different like tasking Claude Code write some scripts or build an application. (Well, unless loopy123 was writing code to delete emails, improbable in this case.).Do you think it's smart to play with tools that you don't understand?
Meta Security Researcher's AI Agent Accidentally Deleted Her Emails
Meta's Summer Yue says she ran OpenClaw on her inbox, but its size 'triggered compaction [and] lost my original instruction' to get her permission before deleting.www.pcmag.com
The process of using new tools is how you gain experience and become proficient in them. So maybe it's not smart to "play" but it could be quite intelligent to incorporate new AI tools in a systematic and careful manner.
you can use the playground on groq.com and try out common ones like gpt-oss 120b or llama 3.3 70b. That will give you a feel for that quality of a few models will put out, for free. You can see how the different models succeed with your research questions. Go there (or another site for free) and enter some halfway through out prompt like this (this is not my field of research, btw):
"Retrofitting homes in the US to include BESS and solar PV has increased as the price of panels has fallen. My small business needs to bid competitively for residential customers to contract for a solar PV and whole home battery backup system. To make my bids competitive, I need to accurately estimate component prices. What 2 material components drive cost the most during this type of installation? How can I obtain these material components at the lowest cost? Reference at least 3 peer reviewed articles published within the last 24 months."
The reason to run local would be now your research questions are not being fed into someone else's system and you can be assured that you will control your research until you are able to publish it. Or if you are researching sensitive areas and don't want to disseminate that content.
If you don't have these concerns or your research is of a more general nature, it is not clear that spending $$$ on you local setup would be worth it, since you could just pay per token with larger, faster models from a cloud service.
Your university may also have a secure AI hub, as does mine.
SpanishAppleNerd
macrumors 6502
Yeah I do want to maintain my research confidential until publishing, the topic is not sensitive but I would like to keep my research for myself and my colleagues for as much time as possible