Apple announced new Mac mini and Mac Studio models, sparking discussion on 5ch’s Business News+ board. The new Mac mini starts at $899, while the top-end Mac Studio configuration runs $5,499, and debate broke out over memory capacity and pricing for running AI locally. Some commented, “You could buy a Hiace (Toyota’s popular commercial van) for what the Mac Studio costs,” and the thread expanded into comparisons with Windows.
(Bloomberg): Apple Inc. unveiled the latest models of its “Mac mini” and “Mac Studio” desktop computers on the 25th, significantly upgrading the processors in both high-demand machines.
The new Mac mini starts at $899 and comes equipped with Apple’s new “M6” chip or the “M5 Pro.” For the Mac Studio, the M5 Max configuration starts at $2,499, while the new M5 Ultra configuration runs $5,499.
According to Apple, both models are available for pre-order starting the 25th, with in-store sales beginning September 22. They will ship with the latest version of the Mac operating system, “macOS 27 Golden Gate,” pre-installed. The OS will begin rolling out to other Mac users this fall.
Source: news.yahoo.co.jp / Original article here
If the Republicans somehow win the midterms, the world will probably start shifting away from US reliance, toward Medusa Halo + Linux Mint or OpenSUSE.
Calling one model choice a "necessary evil" — you don't understand the world at all.
Amazing you can write about things with such a narrow worldview.
Not that I'm buying, lol
But if the new one is 4x faster at AI, guess it's fine I never got my hands on the old model.
If you're aiming for 100B, you need 128GB.
DDR5-44800 64GB x2 runs about ¥300,000, you know?
It's unified memory, but the mini tops out at 64GB while the Studio goes up to 512GB, so it should work fine either way?
Configure the Studio with 128GB RAM and a 1TB SSD and it ends up costing about the same as an NVIDIA DGX Spark.
My old Catalina-era iMac still works fine for most things, but it's no good for AI.
At this rate it's not even going to fit on a phone. Not good enough yet.
Totally different thing — that's comparing to an Intel-era model. The original was a PowerPC, ancient history now.
Then it went Intel.
Then M1.
Completely different machines each time.
That's true for most people.
But if you want an AI agent actually doing work for you,
running it yourself keeps costs down.
If you don't get that, you don't need this at all.
Claude Code
OpenAI Codex
Peter Schneider, who made "Clawd Code," has joined OpenAI and says he's building something even moms can use, so before long
something that does this remotely — and cheaper, regardless of the big names — should show up.
Enjoy the wait until then.
If it's not for work, this is otaku-level hobby stuff.
Depends on what you actually want to do with it.
If you don't even know that, you don't need either one.
Windows, on the other hand…
She doesn't seem to have any complaints.
They claim 20% of Americans use Linux — yeah, sure.
Meanwhile in the People's Republic of China there are orders to use Linux.
What's that about?
It's been hot this year, so I figured whatever runs power-efficiently is fine — that's how I ended up here.
Given the power of this thing and the workloads it's meant to handle, I still can't wrap my head around that amount of memory being enough.
It'll end up as an expensive paperweight.
An iPhone, sure, if it's cheap, go ahead and buy one.
Exactly right.
Switching over ends up cheaper in total cost.
Software assets? That's rich coming from Mac, which has abandoned its own software every time it switched architectures.
With Windows, apps from 20 years ago still run, and you've got virtualization too.
Mac's CPU architecture has changed three times, after all.
Meanwhile some Windows 95-era software still runs on Windows — actually, Windows's backward compatibility is just absurd.
If they can both do the same thing, just buy whichever one you like, no?
Background and Key Points
The Mac mini and Mac Studio sit at opposite ends of Apple’s desktop line, and both now run on Apple Silicon’s unified memory architecture, where CPU and GPU draw from the same pool of fast RAM rather than the separate system-RAM/VRAM split found in Windows PCs. That distinction is why the thread fixated on memory ceilings: unified memory capacity is effectively the ceiling on how large a language model you can run locally, and the Mac Studio’s jump to 512GB (versus the mini’s 64GB) is what makes it a plausible alternative to Nvidia’s DGX Spark for local AI work. The yen angle also needs unpacking for non-Japanese readers — the joke about “¥100 to the dollar” nods to an exchange rate last seen around 2012-2015; at today’s much weaker yen, Apple’s dollar list prices translate into a far heavier local price tag, which is also why a Toyota HiAce commercial van (several million yen) became the go-to comparison for “absurdly expensive.”
The thread’s real fault line wasn’t Mac versus Windows in the abstract, but whether paying Apple’s premium for local inference is worth it at all when cloud-based coding agents (Claude Code, OpenAI Codex) already exist — one camp argued local hardware only pays off if you’re running an always-on agent, the other insisted cloud remains cheaper for nearly everyone.
What the thread never addressed is Apple’s own stated memory guidance for Apple Intelligence, which is separate from and can compound with whatever headroom a local LLM needs — meaning the base 16GB mini configuration posters worried about may already be undersized before any AI workload is loaded.
*This article is compiled from excerpts and a summary of the 5ch (Business News+) thread “[PC] Apple Refreshes Mac Mini and Mac Studio.”
