Apple introduces M6 and M5 Ultra for a big leap in performance and AI compute

M6, Apple’s first 2 nm chip, features a larger, much powerful 12-core CPU, 12-core GPU, and Dual 16-core Neural Engine, while M5 Ultra is Apple’s first quad-die architecture and its astir powerful spot ever

CUPERTINO, CALIFORNIA Apple coming debuted M6 successful the caller Mac mini and M5 Ultra successful the caller Mac Studio, providing an bonzer leap successful capacity and AI capabilities. M6, Apple’s first state-of-the-art 2-nanometer chip, advances each compute block, delivering gains crossed each magnitude of performance. The spot features a larger 12-core CPU analyzable pinch the world’s fastest CPU core, a larger 12-core GPU pinch Neural Accelerators, a Dual 16-core Neural Engine, and up to 170GB/s of unified representation bandwidth.1 M5 Ultra, the eventual powerhouse for pro and AI workloads, uses next-generation UltraFusion exertion to shape a quad-die architecture for the first clip successful an M-series strategy connected a spot (SoC). The spot includes an up-to-36-core CPU and up-to-80-core GPU pinch a monolithic 1.2TB/s of unified representation bandwidth, 50 percent much than M3 Ultra. With their precocious technologies, these SoCs present bonzer compute pinch industry-leading powerfulness efficiency, empowering users to do moreover much connected a desktop.

“Today, we’re debuting the adjacent elephantine leap successful capacity and AI compute for Apple silicon pinch the incredibly precocious M6 and the astir powerful M-series spot yet, M5 Ultra,” said Sri Santhanam, Apple’s vice president of Silicon Engineering Group. “Built utilizing the cutting-edge 2 nm process, M6 combines a caller CPU complex, 2 further CPU and GPU cores, a Dual 16-core Neural Engine, and much unified representation bandwidth to powerfulness done workloads pinch astonishing power efficiency. And for the eventual desktop capacity and the expertise to tally monolithic AI models, M5 Ultra features a monolithic GPU, now pinch Neural Accelerators, and much unified representation bandwidth, pushing the boundaries of what a desktop tin do.”

M6: Optimized Design and Enhanced Performance

Designed to powerfulness the workflows of mundane users, students, developers, AI hobbyists, and enterprises, M6 offers the perfect equilibrium of performance, powerfulness efficiency, and on-device AI to effortlessly alert done regular tasks, coding, and imaginative projects.

M6 is built utilizing cutting-edge 2 nm process technology, packing greater transistor density into a smaller dice for a awesome leap successful capacity and powerfulness efficiency. M6 besides introduces a Dual 16-core Neural Engine, providing up to 2x the highest compute complete erstwhile generations to make on-device AI workflows tally moreover faster. System frameworks tin automatically utilize some engines simultaneously, enabling applications to spot faster exemplary execution.

M6 has a brand-new 12-core CPU analyzable — 2 much cores than M5 — that consists of 2 ace cores, 4 capacity cores, and 6 ratio cores. It delivers the world’s fastest single-threaded capacity and up to 1.2x faster multithreaded capacity arsenic compared to M5, and up to 2.4x faster than M1.2 The ace cores blaze done single-threaded workloads, the capacity cores usage little powerfulness and subordinate the ace cores to tally demanding multithreaded workloads, and the ratio cores grip mundane inheritance tasks — each pinch industry-leading capacity per watt. As a result, demanding CPU tasks specified arsenic editing images, compiling code, indexing caller files, and moving agentic AI workloads are faster than ever.

Accelerated GPU and Faster Memory Bandwidth

M6 features a 12-core GPU — 2 much cores than M5 — pinch a Neural Accelerator successful each core. This creation delivers a astir 30 percent summation successful highest GPU compute for AI compared to M5, and much than 8x compared to M1, enabling importantly faster punctual processing erstwhile interacting pinch on-device LLMs.2

In addition, M6 offers Apple’s latest precocious graphics capabilities, including updates to the shader halfway architecture, Dynamic Caching, and hardware-accelerated ray tracing. These technologies harvester to present stunning ocular realism, faster rendering, and higher framework rates for gaming. M6 besides has 50 percent accrued geometry rates for analyzable graphics.

M6 supports up to 32GB of unified representation to multitask crossed demanding apps and tally LLMs connected instrumentality for unafraid and backstage agentic tasks. It besides provides up to 170GB/s of unified representation bandwidth — a 10 percent summation complete M5 and a 2.5x summation complete M1.

M5 Ultra: The Ultimate Powerhouse for Pro Workflows

M5 Ultra, Apple’s astir powerful spot ever, is built for pros who request to velocity done workloads that request maximum CPU and GPU capacity and unified representation bandwidth, specified arsenic analyzable 3D rendering, ocular effects, technological analysis, and moving compute-intensive frontier AI models connected device.

M5 Ultra uses UltraFusion to link 2 dual-die M5 Max chips to shape the quad-die architecture — a first for Apple silicon. UltraFusion increases the inter-die bandwidth to complete 4.4TB/s and the relationship density by complete 6x. Together, these ultra-low-latency, high-bandwidth interconnects let the 4 dies to behave arsenic a azygous unified processor. M5 Ultra besides features a ample up-to-36-core CPU consisting of 12 ace cores and 24 capacity cores, delivering up to 1.25x higher single-threaded capacity and up to 1.3x higher multithreaded capacity than M3 Ultra.3

Unprecedented AI and Graphics, Massive Memory Capacity

M5 Ultra features a next-generation GPU pinch up to 80 cores, incorporating a Neural Accelerator successful each halfway to connection up to 4.5x the highest GPU compute for AI compared to M3 Ultra and complete 6x much than M1 Ultra.3 The GPU includes Apple’s latest shader halfway pinch second-generation Dynamic Caching, arsenic good arsenic hardware-accelerated mesh shading and third-generation ray tracing, delivering up to 40 percent faster graphics capacity than M3 Ultra.3

M5 Ultra incorporates a much tin Media Engine. Dedicated hardware-enabled H.264, HEVC, 4 ProRes encode and decode engines, and hardware-accelerated AV1 decode make it the eventual solution for high-resolution video editing. M5 Ultra besides includes a 32-core Neural Engine, driving analyzable AI tasks and Apple Intelligence features securely connected instrumentality pinch industry-leading power efficiency.4

Additionally, M5 Ultra features a monolithic magnitude of high-bandwidth unified memory, up to 512GB, and delivers a staggering 1.2TB/s of unified representation bandwidth that is 50 percent higher than M3 Ultra. This lets users shop immense datasets wholly successful section memory, summation the tokens-per-second speed, and tally immense LLMs pinch hundreds of billions of parameters wholly connected device.

Unleashing Power for Developers

Apple’s developer frameworks and devices — including Core AI, Core ML, Metal, and Xcode — pat straight into the precocious hardware of some chips. Developers tin leverage the Dual 16-core Neural Engine successful M6, and the Neural Accelerators successful the GPU pinch the monolithic 512GB unified representation excavation and faster 1.2TB/s of unified representation bandwidth successful M5 Ultra, to supply unthinkable AI compute capabilities.

With these frameworks and caller chips, developers tin tally and fine-tune ample AI models locally connected their Mac. Apple’s developer devices and frameworks automatically optimize capacity crossed the CPU, GPU, and Neural Engine, and springiness developers the expertise to usage Apple Foundation Models, App Intents to pat into Apple Intelligence features, aliases their ain proprietary AI models to build and tally powerful AI workloads wholly connected device.

1. Testing was conducted by Apple successful August 2026 utilizing shipping competitory systems and prime industry-standard benchmarks.
2. Testing was conducted by Apple successful August 2026 utilizing preproduction Mac mini pinch M6 pinch 12-core CPU, 12-core GPU, and 32GB of memory; 14-inch MacBook Pro pinch M5 pinch 10-core CPU, 10-core GPU, and 32GB of memory; and Mac mini pinch M1 pinch 8-core CPU, 8-core GPU, and 16GB of memory. Performance was measured utilizing prime industry‑standard benchmarks. Performance tests were conducted utilizing circumstantial machine systems and bespeak the approximate capacity of Mac mini. See apple.com/mac-mini for much information.
3. Testing was conducted by Apple successful August 2026 utilizing preproduction Mac Studio pinch M5 Ultra pinch 36-core CPU, 80-core GPU, and 256GB of memory; Mac Studio pinch M3 Ultra pinch 32-core CPU, 80-core GPU, and 256GB of memory; and Mac Studio pinch M1 Ultra pinch 20-core CPU, 64-core GPU, and 64GB of memory. Performance was measured utilizing prime industry‑standard benchmarks. Performance tests were conducted utilizing circumstantial machine systems and bespeak the approximate capacity of Mac Studio. See apple.com/mac-studio for much information.
4. Apple Intelligence features are presently disposable for testing done the Apple Beta Software Program, and will beryllium disposable pinch macOS 27 this autumn for users pinch an Apple Intelligence-enabled instrumentality group to a supported language. Apple Intelligence is disposable pinch support for these languages: English, Danish, Dutch, French, German, Italian, Norwegian, Portuguese, Spanish, Swedish, Turkish, Vietnamese, Chinese (simplified), Chinese (traditional), Japanese, and Korean. Some features whitethorn not beryllium disposable successful each regions aliases languages. For characteristic and connection readiness and strategy requirements, spot apple.com/apple-intelligence.