[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"article-mips-risc-v-ai-ip-ces-edge-models-zh":3,"article-related-mips-risc-v-ai-ip-ces-edge-models-zh":33,"series-model-release-0392d382-6364-45bc-8532-8e6759930499":86},{"id":4,"slug":5,"title":6,"content":7,"summary":8,"source":9,"source_url":10,"author":11,"image_url":12,"cover_image":12,"category":13,"language":14,"translated_content":11,"related_article_id":15,"keywords":16,"key_takeaways":25,"views":29,"created_at":30,"published_at":31,"topic_cluster_id":32},"0392d382-6364-45bc-8532-8e6759930499","mips-risc-v-ai-ip-ces-edge-models-zh","MIPS 推出 RISC-V 邊緣 AI IP","\u003Cp data-speakable=\"summary\">MIPS 發表 S8200，這是一組給邊緣 AI 模型用的 \u003Ca href=\"\u002Ftag\u002Frisc-v\">RISC-V\u003C\u002Fa> 處理器 IP。\u003C\u002Fp>\u003Cp>說真的，這次不是單純秀規格。\u003Ca href=\"https:\u002F\u002Fwww.mips.com\" target=\"_blank\" rel=\"noopener\">MIPS\u003C\u002Fa> 在 CES \u003Ca href=\"\u002Fnews\u002Fwhy-prompt-engineering-is-wrong-about-2026-zh\">2026\u003C\u002Fa> 端出 \u003Cstrong>S8200\u003C\u002Fstrong>，目標很明確，就是把 Transformer 和 \u003Ca href=\"\u002Ftag\u002Fagentic-ai\">agentic AI\u003C\u002Fa> 推到裝置端。它主打的是 IP，不是成品晶片，客戶要自己拿去做 SoC。\u003C\u002Fp>\u003Cp>官方講法也很直接。這個設計把 AI engine 和 RISC-V application core 綁在一起。性能範圍從數十到數百 TOPS，靠 coherent cluster tiling 往上疊。另一個重點是，\u003Ca href=\"https:\u002F\u002Fwww.forwardedgeasic.com\" target=\"_blank\" rel=\"noopener\">ForwardEdge ASIC\u003C\u002Fa> 已經選了這套 IP。這家是 \u003Ca href=\"https:\u002F\u002Fwww.lockheedmartin.com\" target=\"_blank\" rel=\"noopener\">Lockheed Martin\u003C\u002Fa> 旗下公司，做的是自主\u003Ca href=\"\u002Fnews\u002Fai-slop-flooding-music-streaming-apps-zh\">平台\u003C\u002Fa>晶片。\u003C\u002Fp>\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>項目\u003C\u002Fth>\u003Cth>內容\u003C\u002Fth>\u003C\u002Ftr>\u003C\u002Fthead>\u003Ctbody>\u003Ctr>\u003Ctd>產品\u003C\u002Ftd>\u003Ctd>MIPS S8200\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>工作負載\u003C\u002Ftd>\u003Ctd>Transformer 與 agentic AI\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>性能範圍\u003C\u002Ftd>\u003Ctd>數十到數百 TOPS\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>首批矽晶參考平台\u003C\u002Ftd>\u003Ctd>2027 年\u003C\u002Ftd>\u003C\u002Ftr>\u003Ctr>\u003Ctd>已公開客戶\u003C\u002Ftd>\u003Ctd>ForwardEdge ASIC\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftbody>\u003C\u002Ftable>\u003Ch2>MIPS 到底在賣什麼\u003C\u002Fh2>\u003Cp>先講白了，MIPS 不是在賣一顆能直接插電用的晶片。它賣的是 IP。這代表別人可以授權後，塞進自己的晶片設計裡。對邊緣 AI 來說，這種模式很常見，因為大家都想要更低功耗，也想保留客製空間。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780668189122-x05o.png\" alt=\"MIPS 推出 RISC-V 邊緣 AI IP\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>S8200 的架構重點，是把 AI engine 跟 RISC-V 核心放在一起。前者處理矩陣和向量運算，後者處理控制流程和應用邏輯。這種分工很務實，因為 AI 推論不是只有算力，還有資料搬運、排程、快取命中率這些麻煩事。\u003C\u002Fp>\u003Cp>它還支援 \u003Ca href=\"https:\u002F\u002Fpytorch.org\" target=\"_blank\" rel=\"noopener\">PyTorch\u003C\u002Fa> 和 \u003Ca href=\"https:\u002F\u002Fwww.tensorflow.org\" target=\"_blank\" rel=\"noopener\">TensorFlow\u003C\u002Fa>。這點我覺得很重要。硬體再猛，軟體團隊如果要重寫一堆流程，大家只會翻白眼。能接現有框架，導入成本才比較像樣。\u003C\u002Fp>\u003Cul>\u003Cli>RISC-V 核心負責控制與應用邏輯。\u003C\u002Fli>\u003Cli>AI engine 負責向量與矩陣運算。\u003C\u002Fli>\u003Cli>coherent cluster tiling 用來往上堆性能。\u003C\u002Fli>\u003Cli>目標是邊緣推論，不是雲端訓練。\u003C\u002Fli>\u003C\u002Ful>\u003Ch2>為什麼 RISC-V 這條路重要\u003C\u002Fh2>\u003Cp>RISC-V 這幾年一直被晶片圈拿來當自由度更高的選項。原因很簡單。大家想少付授權費，也想自己掌控指令集和系統設計。對 AI 晶片來說，這個誘因更明顯，因為很多公司都想把 CPU、NPU、記憶體控制器一起調到最順。\u003C\u002Fp>\u003Cp>但別把 RISC-V 想得太神。它不是萬靈丹。真正難的地方還是軟體、功耗、記憶體頻寬，還有工具鏈。演算法跑得動，不代表裝置端就能穩定量產。這也是為什麼 IP 供應商要連 runtime、編譯器、模型轉換一起想。\u003C\u002Fp>\u003Cp>MIPS 以前就常講 heterogeneous computing。這次 S8200 其實也沿用同一套思路。不是要做一顆什麼都包的通用晶片，而是做給特定工作負載用的積木。對汽車、工業、航太、機器人這些市場，這種做法比較合理。\u003C\u002Fp>\u003Cblockquote>“We believe the future of computing is heterogeneous,” said MIPS CEO Sameer Wasson in a 2024 company statement about the Atlas platform. “We are enabling our customers to co-design solutions that are optimized for their specific workloads.”\u003C\u002Fblockquote>\u003Cp>這段話放到 S8200 身上也很貼。MIPS 的打法一直沒變，就是賣可組合的積木。它不跟 \u003Ca href=\"\u002Ftag\u002Fnvidia\">NVIDIA\u003C\u002Fa> 那種完整平台硬碰硬，而是去搶那些要自己做晶片的客戶。講白了，就是賣刀具，不賣便當。\u003C\u002Fp>\u003Ch2>跟其他邊緣 AI 方案怎麼比\u003C\u002Fh2>\u003Cp>MIPS 這次沒有丟 \u003Ca href=\"\u002Ftag\u002Fbenchmark\">benchmark\u003C\u002Fa> 分數。只有 TOPS 範圍，所以比較時要老實一點。真正的問題不是誰數字最大，而是誰能把功耗、成本、軟體和時程一起壓住。邊緣 AI 很現實，裝置端沒辦法像\u003Ca href=\"\u002Ftag\u002F資料中心\">資料中心\u003C\u002Fa>那樣燒電。\u003C\u002Fp>\n\u003Cfigure class=\"my-6\">\u003Cimg src=\"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780668179677-aqnd.png\" alt=\"MIPS 推出 RISC-V 邊緣 AI IP\" class=\"rounded-xl w-full\" loading=\"lazy\" \u002F>\u003C\u002Ffigure>\n\u003Cp>如果拿市場上常見方案來看，S8200 的定位比較像 IP 平台。它跟 \u003Ca href=\"https:\u002F\u002Fwww.nvidia.com\" target=\"_blank\" rel=\"noopener\">NVIDIA\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fwww.qualcomm.com\" target=\"_blank\" rel=\"noopener\">Qualcomm\u003C\u002Fa>、\u003Ca href=\"https:\u002F\u002Fwww.amba.com\" target=\"_blank\" rel=\"noopener\">Ambarella\u003C\u002Fa> 這些做法不太一樣。那些公司多半賣的是完整晶片或模組，生態比較成熟，但客製彈性通常沒那麼大。\u003C\u002Fp>\u003Cp>這裡可以直接拆成幾種路線來看。每條路都有代價，也都有市場。\u003C\u002Fp>\u003Cul>\u003Cli>\u003Cstrong>MIPS S8200\u003C\u002Fstrong>：IP 模式，適合自研晶片團隊。\u003C\u002Fli>\u003Cli>\u003Cstrong>GPU 型邊緣方案\u003C\u002Fstrong>：生態強，但客製彈性較低。\u003C\u002Fli>\u003Cli>\u003Cstrong>視覺導向加速器\u003C\u002Fstrong>：效率高，但模型範圍常較窄。\u003C\u002Fli>\u003Cli>\u003Cstrong>通用 NPU\u003C\u002Fstrong>：整合容易，但不一定夠好調。\u003C\u002Fli>\u003C\u002Ful>\u003Cp>公開客戶是 \u003Ca href=\"https:\u002F\u002Fwww.forwardedgeasic.com\" target=\"_blank\" rel=\"noopener\">ForwardEdge ASIC\u003C\u002Fa>，這點很有意思。自主平台通常對延遲、可靠性、功耗都很龜毛。這類案子不會只看峰值 TOPS，還會看每瓦性能、記憶體行為，還有系統整合難度。\u003C\u002Fp>\u003Cp>MIPS 說第一批 S8200 矽晶參考平台會在 2027 年出來。時間不短，但 IP 本來就不是快消品。從授權到 tape-out，再到量產，中間常常要磨很久。CES 很多 demo 聽起來都很猛，最後能不能進\u003Ca href=\"\u002Fnews\u002F60305-rule-editing-first-ai-products-zh\">產品\u003C\u002Fa>線，才是真的。\u003C\u002Fp>\u003Ch2>這組資料透露的產業脈絡\u003C\u002Fh2>\u003Cp>邊緣 AI 的方向很清楚。大家都想把推論留在裝置端。原因不是情懷，是錢。雲端推論有延遲，也有持續的伺服器成本。裝置端如果能自己跑，就能省頻寬，也能少依賴網路。\u003C\u002Fp>\u003Cp>這波需求會先落在幾個地方。車用、工控、機器人、監控、國防系統，都很需要低功耗推論。這些場景不會只問模型準不準，還會問斷網能不能跑，溫度高不高，供電穩不穩。這也是為什麼 IP 供應商最近特別愛講 heterogeneous architecture。\u003C\u002Fp>\u003Cp>我覺得 S8200 的價值，不在於它是不是最強。它的價值在於，它把 RISC-V 和 AI engine 的組合講得很明白。對想做自家晶片的團隊來說，這比空喊 AI 更實際。畢竟最後要交付的是產品，不是簡報。\u003C\u002Fp>\u003Ch2>接下來該看什麼\u003C\u002Fh2>\u003Cp>接下來最重要的，是軟體細節。MIPS 如果願意多講 compiler、runtime、memory hierarchy、model conversion，這顆 IP 才會更容易判斷。沒有這些資訊，大家只能看架構圖，還很難知道實戰成績。\u003C\u002Fp>\u003Cp>另一個重點是每瓦性能。S8200 說自己能做到數十到數百 TOPS，但邊緣 AI 真正的戰場是 watts。10 TOPS 如果只吃 2W，跟 100 TOPS 但吃 50W，根本不是同一種產品。這種差距，工程師一看就懂。\u003C\u002Fp>\u003Cp>我的判斷很直接。MIPS 這次是在押注一件事：RISC-V 加上 AI 專用 IP，會比固定功能晶片更適合邊緣推論。這個賭法不算新，但執行起來很硬。你如果是做晶片或系統整合，接下來要盯的不是新聞稿，而是工具鏈和樣板晶片能不能準時出來。\u003C\u002Fp>","MIPS 在 CES 發表 S8200，主打 RISC-V 邊緣 AI IP，鎖定 Transformer 與 agentic 模型，性能範圍從數十到數百 TOPS，首批矽晶參考平台預計 2027 年。","www.electronicsweekly.com","https:\u002F\u002Fwww.electronicsweekly.com\u002Fnews\u002Fdesign\u002Feda-and-ip\u002Fces-risc-v-ai-neural-processor-ip-from-mips-2026-01\u002F",null,"https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780668189122-x05o.png","model-release","zh","f9d8df2e-11f9-45cb-8924-b87d697db555",[17,18,19,20,21,22,23,24],"MIPS","RISC-V","邊緣 AI","S8200","TOPS","Transformer","agentic AI","晶片 IP",[26,27,28],"MIPS S8200 是給邊緣 AI 用的 RISC-V 處理器 IP，不是成品晶片。","官方主打數十到數百 TOPS，首批矽晶參考平台預計 2027 年。","ForwardEdge ASIC 已選用這套 IP，顯示它已進入實際設計案。",2,"2026-06-05T14:02:32.582526+00:00","2026-06-05T14:02:32.549+00:00","0ccb5d2e-69f1-4354-a3e0-cb370221cd95",{"tags":34,"relatedLang":45,"relatedPosts":49},[35,37,39,41,43],{"name":19,"slug":36},"邊緣-ai",{"name":20,"slug":38},"s8200",{"name":21,"slug":40},"tops",{"name":18,"slug":42},"risc-v",{"name":17,"slug":44},"mips",{"id":15,"slug":46,"title":47,"language":48},"mips-risc-v-ai-ip-ces-edge-models-en","MIPS shows RISC-V AI IP for edge models at CES","en",[50,56,62,68,74,80],{"id":51,"slug":52,"title":53,"cover_image":54,"image_url":54,"created_at":55,"category":13},"003f59ba-8d5a-40cb-8e6b-0b51898bc537","midjourney-21-second-video-model-closed-ai-wrong-deal-zh","為什麼 Midjourney 的 21 秒影片模型證明封閉式 AI 是錯的交易","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780648396657-oa18.png","2026-06-05T08:32:37.400103+00:00",{"id":57,"slug":58,"title":59,"cover_image":60,"image_url":60,"created_at":61,"category":13},"b5889da8-fa42-44ed-89a7-3347655b388d","microsoft-seven-ai-models-openai-anthropic-build-2026-zh","7 款 Microsoft AI 模型登場","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780642975404-f8mr.png","2026-06-05T07:02:23.607092+00:00",{"id":63,"slug":64,"title":65,"cover_image":66,"image_url":66,"created_at":67,"category":13},"b5926931-ce20-4b9d-8814-a3c960187209","what-we-know-about-gpt-56-release-date-zh","GPT-5.6 何時發布？目前線索整理","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780574585815-dzo7.png","2026-06-04T12:02:35.122398+00:00",{"id":69,"slug":70,"title":71,"cover_image":72,"image_url":72,"created_at":73,"category":13},"1985ce38-03c6-4968-96fa-b751553bbef3","why-claude-opus-48-is-not-the-big-story-zh","為什麼 Claude Opus 4.8 不是大新聞","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780531367297-nrfs.png","2026-06-04T00:02:24.633987+00:00",{"id":75,"slug":76,"title":77,"cover_image":78,"image_url":78,"created_at":79,"category":13},"8810b91a-9aa2-4cd6-a58b-18fad5897423","devin-booker-sedona-mcdonalds-shoe-launch-zh","Booker把Sedona麥當勞變鞋款發表場","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780510686292-fm1k.png","2026-06-03T18:17:31.966783+00:00",{"id":81,"slug":82,"title":83,"cover_image":84,"image_url":84,"created_at":85,"category":13},"d4d7e664-cc7f-4211-a733-b7c111b86bd6","best-open-source-llms-2026-ranked-zh","2026 最佳開源 LLM 排名","https:\u002F\u002Fxxdpdyhzhpamafnrdkyq.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Fcovers\u002Finline-1780396385004-yyka.png","2026-06-02T10:32:37.264398+00:00",[87,92,97,102,107,112,117,122,127,132],{"id":88,"slug":89,"title":90,"created_at":91},"58b64033-7eb6-49b9-9aab-01cf8ae1b2f2","nvidia-rubin-six-chips-one-ai-supercomputer-zh","NVIDIA Rubin 把六顆晶片塞進 AI 機櫃","2026-03-26T07:18:45.861277+00:00",{"id":93,"slug":94,"title":95,"created_at":96},"0dcc2c61-c2a6-480d-adb8-dd225fc68914","march-2026-ai-model-news-what-mattered-zh","2026 年 3 月 AI 模型新聞重點","2026-03-26T07:32:08.386348+00:00",{"id":98,"slug":99,"title":100,"created_at":101},"214ab08b-5ce5-4b5c-8b72-47619d8675dd","why-small-models-are-winning-on-device-ai-zh","小模型為何吃下裝置端 AI","2026-03-26T07:36:30.488966+00:00",{"id":103,"slug":104,"title":105,"created_at":106},"785624b2-0355-4b82-adc3-de5e45eecd88","midjourney-v8-faster-images-higher-costs-zh","Midjourney V8 變快了，也變貴了","2026-03-26T07:52:03.562971+00:00",{"id":108,"slug":109,"title":110,"created_at":111},"cda76b92-d209-4134-86c1-a60f5bc7b128","xiaomi-mimo-trio-agents-robots-voice-zh","小米 MiMo 三模型瞄準代理、機器人與語音","2026-03-28T03:05:08.779489+00:00",{"id":113,"slug":114,"title":115,"created_at":116},"9e1044b4-946d-47fe-9e2a-c2ee032e1164","xiaomi-mimo-v2-pro-1t-moe-agents-zh","小米 MiMo-V2-Pro 登場：1T MoE 模型","2026-03-28T03:06:19.002353+00:00",{"id":118,"slug":119,"title":120,"created_at":121},"c4b6186f-bd84-4598-997e-c6e31d543c0d","cursor-composer-2-agentic-coding-model-zh","Cursor Composer 2 走向代理式寫碼","2026-03-28T03:13:06.422716+00:00",{"id":123,"slug":124,"title":125,"created_at":126},"e112e76f-ec3b-408f-810e-e93ae21a888a","apple-siri-gemini-distilled-models-zh","Apple Siri 牽手 Gemini 的真相","2026-03-29T04:52:57.886544+00:00",{"id":128,"slug":129,"title":130,"created_at":131},"c679b51f-194a-463b-87fc-7695256ff752","mimo-v2-pro-vs-omni-vs-flash-2026-zh","MiMo V2 Pro、Omni、Flash 怎麼選","2026-04-02T01:18:43.576128+00:00",{"id":133,"slug":134,"title":135,"created_at":136},"3b988fd7-6749-4f01-ba25-c0ad7486dc31","z-ai-glm-5v-turbo-design2code-claude-zh","GLM-5V-Turbo 在 Design2Code 贏了…","2026-04-02T04:03:36.31741+00:00"]