Apple debuts 2nm M6 chip for Mac mini and quad-die M5 Ultra for Mac Studio, targeting a jump in AI compute

Apple introduced two new high-end chips for its desktop lineup on Tuesday, positioning the M6 and M5 Ultra as a step-change in both raw computing speed and on-device AI capabilities.

The company said the M6 is its first 2-nanometer chip and will be used in an updated Mac mini. Apple claims the move advances every part of its compute stack, with a 12-core CPU complex, a 12-core GPU and a Dual 16-core Neural Engine, plus up to 170GB/s of unified memory bandwidth. Apple also said the GPU design includes neural accelerators in each core to speed AI workloads, particularly when users interact with on-device large language models.

On the CPU side, Apple described a “new 12-core CPU complex” that includes two super cores, four performance cores and six efficiency cores. It said the architecture delivers the world’s fastest single-threaded performance, up to 1.2x faster multithreaded performance than the prior M5 generation, and up to 2.4x faster than M1. Apple also linked those gains to practical tasks such as image editing, code compilation, file indexing and “agentic AI workloads,” while emphasizing power efficiency via performance per watt.

For AI specifically, Apple said the M6’s Dual 16-core Neural Engine supports up to 2x peak compute over previous generations, and that system frameworks can automatically use both engines at the same time. The company said this enables faster model execution across applications rather than requiring developers to manually split work between engines.

Apple also paired the M6 with updated graphics features it said are designed for better visual realism and rendering performance. Those additions include updates to shader core architecture, Dynamic Caching, and hardware-accelerated ray tracing, alongside a claim of 50 percent higher geometry rates for complex graphics. Apple said the overall result is faster gaming frame rates and higher-quality rendering on a desktop-class chip.

For the Mac Studio, Apple introduced the M5 Ultra, which it called its most powerful M-series chip to date and “ultimate powerhouse” for pro workflows and large AI models run locally. The company highlighted a quad-die architecture for the first time in an M-series system on a chip, achieved using its UltraFusion technology to connect two dual-die M5 Max chips.

Apple said UltraFusion increases inter-die bandwidth to over 4.4TB/s and boosts connection density by more than 6x, enabling the four dies to behave as a unified processor with low latency and high bandwidth. It said the chip includes up to a 36-core CPU, built from 12 super cores and 24 performance cores, with performance claims of up to 1.25x higher single-threaded performance and up to 1.3x higher multithreaded performance than M3 Ultra.

On AI and memory, Apple stated that the M5 Ultra features a next-generation GPU with up to 80 cores, again embedding a Neural Accelerator in each core. It claimed up to 4.5x peak GPU compute for AI compared with M3 Ultra and over 6x compared with M1 Ultra. It also said the chip includes a 32-core Neural Engine and can support up to 512GB of unified memory, with 1.2TB/s of unified memory bandwidth, described as 50 percent higher than M3 Ultra. Apple tied those specifications to the ability to keep large datasets locally, increase tokens per second, and run very large LLMs with hundreds of billions of parameters entirely on device.

Apple added that its developer frameworks and tools, including Core AI, Core ML, Metal and Xcode, are designed to tap directly into the underlying hardware of the chips. It said developers can run and fine-tune large AI models locally on Macs, with the tools automatically optimizing across CPU, GPU and Neural Engine. Apple also referenced the ability to use Apple Foundation Models, App Intents for Apple Intelligence features, or developers’ own proprietary models, positioning the hardware update as an enabler for a broader set of on-device AI applications.

A key caveat is what Apple did not provide in its announcement: it did not share benchmark results versus specific competing chips, nor did it detail pricing, configuration differences beyond capacity and throughput claims, or availability timing for the new Mac mini and Mac Studio models. The company also did not specify which Mac Studio configurations pair the M5 Ultra with which memory or whether every memory tier is supported at the stated bandwidth and unified capacity figures. Those details will likely be important for buyers and developers trying to translate chip specifications into real-world performance.

Next, Apple’s own developer and performance materials, including any published benchmarks for the M6 and M5 Ultra, will be the first indicators of how these architectural changes affect typical workloads such as code builds, creative rendering and local LLM inference. Watch also for software updates that explicitly exercise the Dual 16-core Neural Engine on M6 and the quad-die compute and large unified memory capabilities on M5 Ultra, since Apple’s claims repeatedly point to system frameworks and frameworks-level optimizations as the mechanism for extracting the promised AI acceleration.