Apple Unveils M5 Ultra and M6 Chips, Targeting On-Device AI Workloads
Nashnova编辑部
Apple on Tuesday launched two new processors — the M5 Ultra and M6 — powering the new Mac Studio and Mac Mini respectively, both built around a single pitch: run AI models locally on your machine, not in the cloud.
Why does Apple call the M5 Ultra its "most powerful chip ever"?
It fuses two M5 Max dies into one package — Apple's first four-die chip.
Top configuration: 36-core CPU, 80-core GPU, unified memory bandwidth of 1.2 TB/s. This means → the chip moves data 50% faster than the M3 Ultra released roughly eighteen months ago.
In plain terms = this is not a consumer chip. It targets 3D rendering, visual effects, and running large AI models — workloads that devour compute.
Who is M6 for, and how does it improve on M5?
M6 is built on a 2-nanometer process — one of the most advanced nodes in mass production — and adds two CPU cores over M5 for a 12-core total.
The headline number: peak GPU AI performance is up nearly 30% versus M5, which directly speeds up prompt processing for large language models running locally.
Apple pairs it with a dual 16-core Neural Engine — dedicated hardware that accelerates AI computation — and raises memory bandwidth. This means → Apple wants "fine-tuning an AI model on your Mac" to be a routine developer workflow, not a fringe experiment.
Apple is betting on local AI — what is the logic?
Apple trails Google and OpenAI in building its own large AI models — even the long-awaited Siri upgrade is powered by Google's Gemini.
But the company has years of on-device processing expertise: Face ID and Touch ID run entirely on the device, never uploading data. That is its privacy moat.
This reflects a deliberate differentiation play: rather than competing on who has the strongest cloud model, Apple is competing on who can make AI run safely on the user's own hardware.
Can this strategy work? What is the deciding factor?
Apple ships developer tools and frameworks that automatically coordinate CPU, GPU, and Neural Engine resources, supporting Apple's foundation models, App Intents, and developers' proprietary AI models.
In plain terms = Apple has laid out the hardware and the tooling. Whether it pays off hinges on how deeply developers actually adopt on-device AI instead of defaulting to cloud APIs.
The new M6 Mac Mini is available for pre-order at $899, shipping after September 22 — no dramatic price hike. This means → Apple wants local AI to reach mainstream consumers, not just professionals.
Content is for reference only, not financial advice.