Apple's new M6 chip gets more cores and more AI compute 苹果新款M6芯片获得更多核心和更多AI算力

TL;DR

• Apple unveiled the M6, its first 2nm chip, featuring a 12-core CPU, 12-core GPU, and Dual 16-core Neural Engine for enhanced on-device AI workloads
• The M5 Ultra is Apple's most powerful chip ever, combining two dual-die M5 Max chips via UltraFusion into a quad-die architecture with up to 36 CPU cores, 80 GPU cores, and 512GB unified memory
• The M6 delivers the world's fastest single-threaded performance and up to 1.2x faster multithreaded performance compared to the M5
• Both chips are designed to support frontier AI model inference and demanding tasks like 3D rendering directly on-device
• The M6 launches in a new Mac Mini and the M5 Ultra in a new Mac Studio, both available September 22nd

Why It Matters

Apple's aggressive push into 2nm fabrication and quad-die chip design signals a significant escalation in the competitive race for on-device AI performance, reducing reliance on cloud-based inference. The M5 Ultra's 512GB unified memory capacity is particularly notable for researchers and practitioners who need to run large language models locally without external hardware dependencies.

Technical Details

• M6 Architecture: First Apple 2nm chip with a 12-core CPU (2 super cores, 4 performance cores, 6 efficiency cores), 12-core GPU with Neural Accelerator per core, and Dual 16-core Neural Engine; supports up to 32GB unified memory
• M5 Ultra Architecture: Quad-die design using UltraFusion interconnect to combine two dual-die M5 Max chips; up to 36-core CPU (12 super cores, 24 performance cores), up to 80-core GPU, 32-core Neural Engine, and up to 512GB high-bandwidth unified memory
• Performance Claims: M6 offers "world's fastest single-threaded performance" and up to 1.2x faster multithreaded performance versus M5
• Product Lineup: M6 ships in the new Mac Mini; M5 Ultra ships in the new Mac Studio; preorder begins Tuesday with September 22nd release

Industry Insight

Apple's move to 2nm and quad-die packaging demonstrates that silicon-level innovation remains a key differentiator in the AI hardware arms race, particularly for edge inference workloads. The 512GB unified memory ceiling on the M5 Ultra could make it a compelling option for small-scale AI labs and individual researchers seeking to run frontier models without cloud infrastructure. Expect increased pressure on competitors to match Apple's on-device AI capabilities as consumer and professional expectations for local inference continue to rise.

TL;DR

• Apple发布M6（首款2nm工艺）和M5 Ultra两款新芯片，分别搭载于Mac Mini和Mac Studio
• M6采用12核CPU+12核GPU架构，配备双16核神经引擎，支持32GB统一内存
• M5 Ultra通过UltraFusion技术整合双M5 Max芯片形成四die架构，最高36核CPU/80核GPU/32核神经引擎，支持512GB高带宽内存
• 两款产品面向3D渲染和前沿AI模型等高性能计算场景，9月22日上市

为什么值得看

Apple持续在自研芯片架构上推进制程与集成度创新，M6的2nm工艺和M5 Ultra的四die设计为端侧AI推理提供了新的硬件基础。对关注Apple生态和端侧AI部署的从业者而言，这些进展直接影响本地大模型运行能力和开发工具链选择。

技术解析

• M6芯片架构：采用2nm制程工艺，配备双16核神经引擎（共32核），CPU为12核（2超级核心+4性能核心+6能效核心），GPU为12核且每核集成神经加速器，支持最高32GB统一内存。Apple宣称其单线程性能为"全球最快"，多线程性能较M5提升约1.2倍。
• M5 Ultra芯片架构：通过UltraFusion互联技术将两个双die M5 Max芯片整合为四die架构，最高配置36核CPU（12超级核心+24性能核心）、80核GPU和32核神经引擎，支持最高512GB高带宽统一内存（不可升级）。
• 产品发布计划：M6首发于新Mac Mini，M5 Ultra首发于新Mac Studio，均于9月22日正式发售，提前一周开放预购。

行业启示

• 端侧AI算力竞争加剧：Apple通过神经引擎核心数翻倍和统一内存带宽提升，持续强化本地AI推理能力，表明端侧部署大模型已成为硬件厂商的核心差异化方向。
• 先进制程与Chiplet技术并重：M6采用2nm制程追求能效，M5 Ultra采用多die互联方案扩展算力，两种技术路径并行推进，反映高端芯片设计正走向制程微缩与系统级集成并重的阶段。
• 开发者生态影响：统一内存架构和神经引擎优化将持续影响Core ML等开发工具的性能上限，建议关注Apple对主流AI框架的适配进度，以便提前规划本地部署方案。