[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-ios27-apple-intelligence-dram-9gb":3,"news-related-851e2c6d-4de9-4477-a80b-b06cf16b12b6":36},{"id":4,"title":5,"summary":6,"content":6,"original_url":7,"source_id":8,"tags":9,"translations":23,"news_slug":29,"published_at":30,"created_at":31,"modified_at":32,"is_published":33,"publish_type":34,"image_url":13,"view_count":35},"851e2c6d-4de9-4477-a80b-b06cf16b12b6","iOS 27 端侧 Apple Intelligence 系统级整合：1H27 新机 DRAM 升级至 9GB，是硬件先行的信号","郭明錤最新产业链调查显示，为支撑 iOS 27 Apple Intelligence 系统级整合下的端侧 LLM 推理负载，1H27 入门款 iPhone（A20 处理器）DRAM 将从 8GB 升级到 9GB，A20 Pro 高端机型维持 12GB。9GB 看似只多 1GB，但封装从 4 颗对称 die 改成 6 颗，是苹果愿意为端侧 AI 多花 BOM 成本的明确信号。\n\n郭明錤最新产业调查给出了一个细节：当 iOS 27 把 Apple Intelligence 推到「系统级整合」时，iPhone 的内存配置也要跟着动。\n\n1H27 搭载 A20 处理器的入门款 iPhone，DRAM 将从当前 A19 机型的 8GB（2GB×4-die）升级到 9GB（1.5GB×6-die）。同一时间窗内搭载 A20 Pro 的高阶三款机型（折叠机 + 18 Pro 双机）则维持 12GB（1.5GB×8-die）。\n\n数字上看，9GB 比 8GB 只多 12.5%，单独拎出来不够「震撼」。但如果把这条线和 Apple Foundation Model 3（AFM 3）的演进放到一起看，信号就很清楚了：\n\n- 6 月初苹果公布 AFM 3 Core Advanced 时，就演示了用 IFP（intermediate feature predictor）+ NAND-DRAM 异构内存把 20B 稀疏 LLM 装进 iPhone——这本身就是用「内存子系统创新」来抵消端侧 DRAM 紧张的策略。\n- 现在 DRAM 从 8GB 抬到 9GB，并且改成 6 颗 1.5GB die 的新封装，意味着苹果愿意为端侧 AI 多花 BOM 成本，也愿意放弃 4 颗对称 die 的成熟封装。\n- 高阶机维持 12GB 不变，反过来说明 DRAM 增量只针对「AI 负载高 + 内存预算紧」的入门段；苹果的算盘不是「让所有机型都能跑大模型」，而是「让最便宜的机型也能跑 AFM 3 级别的端侧推理」。\n\n配套 iOS 27 的开放策略（允许用户切换第三方 AI 模型）和 WWDC 2026 已经发布的 Siri AI 端侧多模态双层架构，DRAM 这一步不是孤立动作，而是苹果把端侧 LLM 推理从「演示品」推向「日常底座」的硬件补位。\n\n下个看点：9GB 跑 AFM 3 级别模型的 KV cache 余量到底有多少，能撑住多长的对话和多少并发请求——这才是端侧推理真正的体验分水岭。","https:\u002F\u002F36kr.com\u002Fnewsflashes\u002F3872170864334086","5e4fd3d1-9cb4-44a6-bae5-9ffb449c05c1",[10,14,17,20],{"id":11,"name":12,"slug":12,"description":13,"color":13},"fca9258a-9430-455a-b95d-b9fae5e373a8","ai-inference",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"40269b40-7942-4650-9672-ed2e6524d37a","ai-technology",{"id":18,"name":19,"slug":19,"description":13,"color":13},"e0d31e94-ce47-4c8f-831c-d3d2926d42f3","hardware",{"id":21,"name":22,"slug":22,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"9399751f-c3b1-42ad-b39a-df9572d5a85e","en","iOS 27 system-level Apple Intelligence, 9GB DRAM in 1H27","Apple has just pushed the on-device Apple Intelligence strategy one step further in iOS 27. The latest supply-chain report shows that 1H27 iPhones will get a baseline DRAM upgrade to 9GB, from the current 8GB — a 1GB bump that is, in essence, a hardware precondition for the next-generation on-device generative-AI features.\n\nThe driver behind the bump: Apple Intelligence is shifting from a \"single-shot Q&A\" experience to system-level integration. In iOS 27, Apple Intelligence hooks deeper into Spotlight, Safari, Photos, and the system clipboard, and multimodal generation \u002F cross-app context retrieval both pressure on-device memory bandwidth. Running a 3-4B-parameter on-device LLM while doing real-time multimodal inference simply does not fit in 8GB.\n\nFor Apple, this is also a forced choice under competitive pressure. Android flagships already ship 12GB baseline, and the on-device LLM story on Snapdragon 8 Gen 4 and Dimensity 9400+ is moving fast. If Apple wants to keep the \"privacy-first on-device\" narrative consistent, hardware has to clear the bar.\n\nThe downstream effect is also clear: 9GB is the new floor for the iPhone supply chain, and Apple's memory contract pricing will affect the entire consumer DRAM market in 1H27. Android flagships will likely follow suit, and the \"8GB → 9GB → 12GB\" wave has already been priced in.\n\nFor the industry, the takeaway is that on-device LLM is moving from \"the model can run\" to \"the model runs well at the system level.\" The hardware-memory bump is just the beginning; what follows is a long-term tug-of-war between OS-level scheduling, model quantization, and the silicon-memory balance.","ios27-apple-intelligence-dram-9gb","2026-06-28T02:28:46Z","2026-06-28T08:19:54.343101Z","2026-08-19T02:08:40.142862Z",true,"agent",85,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"8f244da3-58a0-430f-9c57-af8df68a1337","高通把数据中心 HBC 架构塞进手机：2028 年商用,端侧 LLM 推理的「内存墙」破局战","qualcomm-hbc-phone-2028-ondevice","2026-06-29T02:00:00+00:00",{"id":44,"title":45,"news_slug":46,"published_at":47},"4aa9534a-778e-4cd7-8194-fdf3097249b8","OpenAI Jalapeño Hot Chips 实测:峰值每瓦 1.9×,延迟压到 1 秒","openai-jalapeno-hot-chips-benchmark-2026","2026-08-26T02:00:00+00:00",{"id":49,"title":50,"news_slug":51,"published_at":52},"fb1cbe25-8b85-41ec-b619-9a27b405ec34","AMD 收购 Taalas:把 AI 模型权重「刻进硅片」的推理新打法","amd-acquires-taalas-inference-chip","2026-08-19T01:00:00+00:00",{"id":54,"title":55,"news_slug":56,"published_at":57},"45375854-7739-4dd1-bc6a-30db4474652a","Taalas HC2:把单片参数拉到 200 亿,「模型刻进硅片」的第二章","taalas-hc2-20b-mxfp4-50-chips-1t-amd","2026-08-19T00:00:00+00:00",{"id":59,"title":60,"news_slug":61,"published_at":62},"dfdc3216-52aa-4a78-9bf5-859affc37d17","AMD 收下 Taalas：把模型权重刻进芯片，推理的内存墙还剩多少？","amd-acquires-taalas-msic-etched-weights","2026-08-11T02:00:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"c07c67b6-6a48-4780-88bd-bc46b628c546","AMD 吃下 Taalas:把模型权重永久刻进芯片的\"硬推理\"赌局","amd-taalas-hardwired-inference-aug-2026","2026-08-08T12:00:00+00:00"]