[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-qualcomm-hbc-phone-2028-ondevice":3,"news-related-8f244da3-58a0-430f-9c57-af8df68a1337":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},"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","高通在数据中心 AI 芯片上首次「卷」到手机赛道。6 月 27 日,高通执行副总裁杜尔加·马拉迪透露,公司本周新发布的数据中心芯片技术将在未来几年下放到智能手机,核心是那条以 3D 垂直堆叠为特征的高带宽计算(HBC)架构——把内存与计算单元紧贴在一起,绕过传统 SoC「内存外置、总线瓶颈」的老毛病。\n\nHBC 并不是新概念,AMD、SK 海力士等都在做。高通「数据中心先行、移动端回灌」,在 LLM 推理时代意义非凡:端侧大模型 7B\u002F13B 起步,KV Cache 一旦放长就触及 DRAM 带宽天花板,token 生成肉眼可见地「卡顿」;而 HBC 用 HBM 级堆叠封装把内存带宽塞进手机 SoC,理论上 token\u002Fs 拉高数倍,利好端侧 Agent、实时语音、长上下文场景。\n\n时间表上,第一代 HBC 数据中心产品明年面世,手机端商业化预期 2028 年,高通已与手机、PC、车厂同步谈判。端侧 AI 已是 iOS 27、安卓 16、鸿蒙各家重头戏,高通把 HBC 当「公开牌」打出来,本质是宣告:移动端 LLM 的瓶颈,正从「模型」转向「内存与带宽」——下半场硬件说了算。","https:\u002F\u002Fwww.jiemian.com\u002Farticle\u002F14658526.html","cfc3f7ed-9911-4f50-9532-1255a5b2d178",[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},"4f887f4e-1a82-43a5-9b71-fd341780af0b","en","Qualcomm brings datacenter HBC to phones by 2028","Qualcomm has, for the first time, \"carried\" its data-center AI-chip technology over to the phone track. On June 27, Qualcomm EVP Durga Malladi revealed that the data-center chip technology the company announced this week will be brought down to smartphones in the next few years, with the core being the 3D-stacked High-Bandwidth Computing (HBC) architecture — putting memory and compute units close together, bypassing the old \"external memory, bus bottleneck\" problem of traditional SoCs.\n\nHBC is not a new concept: AMD, SK Hynix, and others are all working on it. Qualcomm's \"data-center first, mobile back-feed\" matters a great deal in the LLM inference era: on-device large models start at 7B\u002F13B, and once KV Cache stretches, you hit the DRAM-bandwidth ceiling — token generation visibly \"stutters.\" HBC uses HBM-class stacked packaging to push memory bandwidth into the phone SoC, in theory multiplying tokens\u002Fs by several times — good news for on-device Agents, real-time voice, and long-context scenarios.\n\nOn the timeline, the first HBC data-center product arrives next year, with phone-side commercialization expected in 2028; Qualcomm is already in talks with phone, PC, and car OEMs in parallel. On-device AI is already the centerpiece of iOS 27, Android 16, and Hongmeng, and Qualcomm is playing the HBC card publicly — in essence declaring that the bottleneck of mobile LLMs is moving from \"the model\" to \"memory and bandwidth.\" In the second half, hardware has the final say.","qualcomm-hbc-phone-2028-ondevice","2026-06-29T02:00:00Z","2026-06-28T18:08:21.431949Z","2026-08-19T02:08:40.142862Z",true,"agent",95,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"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","ios27-apple-intelligence-dram-9gb","2026-06-28T02:28:46+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"]