Apple launches M6 and M5 Ultra chips in new Mac Mini and Mac Studio

👁 113 reads

Apple has announced updated Mac mini and Mac Studio desktops alongside two new chips - the M6 and the M5 Ultra - with the machines positioned explicitly around local AI inference and software development. The M6, described by Apple as the first 2nm chip in its M-series Mac lineup, features a 12-core CPU combining two “super cores,” four performance cores, and six efficiency cores, making it the first Apple SoC to use all three core types simultaneously. Apple claims the chip delivers up to 40 percent faster multi-threaded CPU performance compared to the M4.

The M6 also carries a 12-core GPU - two more than its immediate predecessor - and unified memory bandwidth of up to 160GB per second. Memory capacity on the M6, however, tops out at 32GB, which limits its usefulness for running large local models.

The more significant announcement for AI workloads is the M5 Ultra, which powers the refreshed Mac Studio. Effectively two M6 Max dies on a single SoC, it provides 36 CPU cores and 80 GPU cores. Apple claims unified memory bandwidth of up to 1.2TB per second and a maximum memory configuration of 512GB - a figure no single mass-market consumer chip currently matches.

Why Developers Are Paying Attention

Neither machine introduces major new features beyond the chip upgrades. The AI angle reflects a use case that emerged after macOS 26.2 shipped last December, when Apple’s release notes confirmed the update enabled “low-latency communication between Thunderbolt 5 hosts for use cases including distributed AI inference using MLX.” MLX is an open-source array framework built to help machine learning workflows exploit the M-series unified memory architecture.

Since then, hobbyists and professional developers have been chaining multiple Mac minis or Mac Studios together over Thunderbolt 5 to run inference on local large language models too large for any single consumer device - an arrangement that competes, on certain workloads, with specialised Nvidia GPU hardware.

The commercial driver is cost. Cloud-based frontier models accessed through tools like Claude Code or Codex carry ongoing API expenses, and many developers are weighing whether running open-weight models locally on high-memory Apple hardware is a more practical long-term option. The M5 Ultra’s 512GB ceiling and claimed 1.2TB/s bandwidth put it in a different category from previous consumer silicon for that purpose, even if the top configuration will remain out of reach for most buyers.

// read moreSuno Launches Studio 2.0 With MIDI, Plugins, Session ChatbotAI ToolsApple explains how new Audio Intelligence features process sound without storing raw audioAI ToolsIBM Launches Granite 4.2 in Three Sizes With 128K ContextNews

Kinetyra