[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-slug-samsung-gaia-ai-pc-pim":3,"news-related-4f957b38-d8f5-446f-805c-062ff268d7ab":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},"4f957b38-d8f5-446f-805c-062ff268d7ab","三星 GAIA 试水 AI PC：把「存算一体」塞进 NPU，准备用 PIM 抢端侧推理","7 月 10 日，韩媒披露三星电子正研发面向 AI PC 的生成式 AI 加速芯片 GAIA：4nm 工艺、搭载 NPU，并设计支持存算一体（PIM）技术，样片已送往联想、惠普做验证，预计 2027 年量产。\n\nGAIA 并非 GPU 推理方案，而是一颗「以存储为中心」的 AI 加速器——把计算贴近内存，配合 PIM 这一能在 DRAM 内部直接做运算的下一代存储，把大模型推理时数据搬运的能耗压到最低。这与堆 GPU 的传统路线形成对比，更接近 Mythic、Ampere 走过的「内存内计算」思路。\n\n从产品定位看，GAIA 是三星把手机 NPU 方案重新封装为 PC 版本，目标是在 Apple Neural Engine、Qualcomm Hexagon、Intel\u002FAMD 端侧 NPU 之外提供「第五种选择」。三星 2012 年曾把 Exynos 塞进 Chromebook，两年后停摆，这次借 AI PC 二次入场。\n\n真正的看点是 PIM 能否在 PC 上真正落地——存储工艺、编译器栈、模型量化生态缺一不可。若 GAIA 能在 7B-13B 端侧模型上跑出「长续航 + 实时响应」的体验，AI PC 才会从营销话术变成日常工具；若还只能跑语音助手、图像增强这种轻负载，AI PC 概念又会被消耗一轮。","https:\u002F\u002Fwww.chosun.com\u002Fenglish\u002Findustry-en\u002F2026\u002F07\u002F09\u002F4USIV3SG5JDBPBIK7UVBJFYJIM\u002F","787b6c8f-0ae7-46a8-a428-864cc1037638",[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},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",{"id":21,"name":22,"slug":22,"description":13,"color":13},"e0d31e94-ce47-4c8f-831c-d3d2926d42f3","hardware",[24],{"id":25,"lang":26,"title":27,"summary":28,"content":13},"ce65ebd4-e529-4f02-ba56-5a0513aa5ec5","en","Samsung GAIA AI PC: compute-in-memory NPUs for the edge","On July 10, Korean media revealed that Samsung Electronics is developing the generative AI accelerator chip GAIA for AI PCs: 4nm process, equipped with NPU, and designed to support Processing-in-Memory (PIM) technology, with sample chips sent to Lenovo and HP for validation, expected to enter mass production in 2027. GAIA is not a GPU inference solution, but an \"storage-centric\" AI accelerator — putting compute close to memory, paired with PIM, the next-generation storage that can perform operations inside DRAM, minimizing the energy cost of data movement during LLM inference. This contrasts with the traditional route of piling on GPUs, and is closer to the \"compute-in-memory\" thinking that Mythic and Ampere have walked. From a product positioning perspective, GAIA is Samsung repackaging its phone NPU solution as a PC version, aiming to provide a \"fifth option\" besides Apple Neural Engine, Qualcomm Hexagon, Intel\u002FAMD on-device NPUs. Samsung back in 2012 put Exynos in a Chromebook, but it shut down two years later; this time it's using AI PC to make a second entry. The real highlight is whether PIM can actually land on PC — storage process, compiler stack, model quantization ecosystem, none can be missing. If GAIA can deliver \"long battery + real-time response\" on 7B-13B on-device models, AI PC will move from marketing talk to daily tool; if it can only run light loads like voice assistants and image enhancement, the AI PC concept will be consumed for another round.","samsung-gaia-ai-pc-pim","2026-07-11T06:00:00Z","2026-07-11T06:08:43.451410Z","2026-08-19T02:08:40.142862Z",true,"agent",90,{"items":37},[38,43,48,53,58,63],{"id":39,"title":40,"news_slug":41,"published_at":42},"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",{"id":44,"title":45,"news_slug":46,"published_at":47},"2434bbc6-4fda-4750-a02d-dd3ca1fe8933","AMD 收购 Taalas:把模型权重刻进芯片,押注推理硬件的\"硬核\"路线","amd-acquires-taalas-hardcore-inference-silicon","2026-08-08T04:00:00+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},"c26cb1e1-d0c0-471d-81a1-79536834a617","AMD 收下 Taalas：把 Llama 权重烧进 ASIC，推理速度把 GPU 甩在身后","amd-acquires-taalas-hardcore-asic-inference","2026-08-17T00:00:00+00:00",{"id":64,"title":65,"news_slug":66,"published_at":67},"1e553217-9229-4e0c-97e8-9ef8dedb5561","HC1 跑 16,960 tokens\u002F秒的背后:Taalas 把模型烧进硅片的架构账本","taalas-hc1-16960-tokens-architecture","2026-08-13T03:00:00+00:00"]