[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"site-home-live-entry-setting":3,"site-header-contests":8,"site-header-camps":49,"study-material-detail-study-66":104},{"enabled":4,"title":5,"statusText":5,"buttonText":5,"campSlug":5,"stage":6,"classroomId":7,"hasClassroom":4},false,"",0,null,{"total":9,"list":10,"pageNum":44,"pageSize":45,"size":9,"startRow":44,"endRow":9,"pages":44,"prePage":6,"nextPage":6,"isFirstPage":46,"isLastPage":46,"hasPreviousPage":4,"hasNextPage":4,"navigatePages":47,"navigatepageNums":48,"navigateFirstPage":44,"navigateLastPage":44},3,[11,25,35],{"slug":12,"title":13,"logo":7,"liveUrl":14,"prize":15,"sponsor":16,"description":17,"organizer":18,"startTime":19,"endTime":20,"registrationDeadline":21,"status":22,"dashboardVisibility":23,"createdAt":24},"spring2026","2026春季启元人工智能大赛","https:\u002F\u002Fwww.bilibili.com\u002Fblackboard\u002Flive\u002Flive-activity-player.html?cid=1822703816&quality=0","200,000","麒麟软件有限公司","“2026春季启元人工智能大赛”是由启元实验室主办的面向全国人工智能领域开发者、科研人员及高校学生的权威赛事。大赛旨在促进与培养具备工程实践能力和创新能力的优秀人工智能领域开发者、推广国产软硬件技术与生态、助力国家人工智能产业高质量发展。\n各参赛选手需在规定的比赛周期内，依据赛事规则选定赛题并完成相应挑战，优胜者将共同角逐总额20万元的赛事奖金池，并且还将有机会获得实习和知名企业优质岗位内推资格。","启元实验室","2026-04-19 16:00:00","2026-07-13 16:00:00","2026-06-16 16:00:00","ongoing","private","2026-04-16 08:00:00",{"slug":26,"title":27,"logo":7,"liveUrl":14,"prize":28,"sponsor":7,"description":29,"organizer":18,"startTime":30,"endTime":31,"registrationDeadline":32,"status":33,"dashboardVisibility":23,"createdAt":34},"autumn2025","2025秋季启元人工智能大赛","150,000","“2025秋季启元人工智能大赛”是由启元实验室主办的面向全国人工智能领域开发者、科研人员及高校学生的权威赛事。大赛旨在促进与培养具备工程实践能力和创新能力的优秀人工智能领域开发者、推广国产软硬件技术与生态、助力国家人工智能产业高质量发展。","2025-11-08 08:00:00","2026-01-13 08:00:00","2025-12-30 08:00:00","ended","2025-11-03 08:00:00",{"slug":36,"title":37,"logo":7,"liveUrl":14,"prize":38,"sponsor":7,"description":39,"organizer":18,"startTime":40,"endTime":41,"registrationDeadline":42,"status":33,"dashboardVisibility":23,"createdAt":43},"summer2025","启元人工智能大赛(系统方向)","80,000","启元人工智能大赛由启元实验室主办，其中系统赛道基于启元实验室九源软件栈的相关开源项目，通过拉动开源社区力量，一方面促进国产智能计算生态的发展，另一方面也为人工智能系统方向的学生及从业者提供一个实践平台。","2025-07-08 08:00:00","2025-08-25 08:00:00","2025-07-22 08:00:00","2025-06-22 08:00:00",1,5,true,8,[44],{"total":45,"list":50,"pageNum":44,"pageSize":45,"size":45,"startRow":44,"endRow":45,"pages":44,"prePage":6,"nextPage":6,"isFirstPage":46,"isLastPage":46,"hasPreviousPage":4,"hasNextPage":4,"navigatePages":47,"navigatepageNums":103,"navigateFirstPage":44,"navigateLastPage":44},[51,62,73,84,94],{"id":52,"slug":53,"title":54,"logo":55,"publicLiveEmbedUrl":14,"description":56,"introduction":57,"organizer":18,"coOrganizer":5,"sponsor":16,"startTime":58,"endTime":59,"status":22,"dashboardVisibility":60,"createdAt":58,"updatedAt":61},"019eeeff-6b00-7d55-a629-efafded3e232","summer2026","2026年夏季InfiniTensor大模型与人工智能系统训练营","https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fuploads\u002F2026-06\u002F019eef7f-7816-7828-9309-79c2c61dd568.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=uiV9kOffa9Xq8DT%2BbErZU0JyQ98%3D","由启元实验室发起的InfiniTensor大模型与人工智能系统训练营，面向开源社区、高校及产业界伙伴开放。活动以「普及AI全栈知识技术、培养工程实践与创新型人才、推广国产软硬件技术生态」为核心目标，助力国家人工智能产业高质量发展。","...","2026-06-22 08:00:00","2026-10-19 08:00:00","public","2026-10-12 08:00:00",{"id":63,"slug":64,"title":65,"logo":66,"publicLiveEmbedUrl":7,"description":67,"introduction":57,"organizer":18,"coOrganizer":5,"sponsor":68,"startTime":69,"endTime":70,"status":33,"dashboardVisibility":23,"createdAt":71,"updatedAt":72},"019af3f3-e06a-7245-aa8d-37dc99a69c73","winter2025","2025年冬季InfiniTensor大模型与人工智能系统训练营","https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fuploads\u002F2025-12\u002F019afc69-b7ca-7805-91fb-605429671c01.png?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=8u%2BaQaeliN4QzF6MZWiB4jHYDCE%3D","由启元实验室发起的InfiniTensor大模型与人工智能系统训练营，面向开源社区、高校及产业界伙伴开放。活动以「普及AI全栈知识技术、培养工程实践与创新型人才、推广国产软硬件技术生态」为核心目标，助力国家人工智能产业高质量发展。 \n","新华三技术有限公司","2025-12-25 08:00:00","2026-03-27 08:00:00","2025-12-06 08:00:00","2026-10-22 20:24:30",{"id":74,"slug":36,"title":75,"logo":76,"publicLiveEmbedUrl":7,"description":77,"introduction":57,"organizer":18,"coOrganizer":5,"sponsor":78,"startTime":79,"endTime":80,"status":81,"dashboardVisibility":23,"createdAt":82,"updatedAt":83},"01980342-208c-7803-b3f3-19e1d31960ba","2025年夏季InfiniTensor大模型与人工智能系统训练营","https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fuploads\u002F2025-07\u002F0198045e-28fe-78f0-95fe-23835ce58395.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=8pXXbnSlNEtVJCJ1e4RzUlqmRn8%3D","InfiniTensor大模型与人工智能系统训练营由启元实验室及清华大学 AI 系统方向团队发起，面向开源社区、高校和产业界，普及人工智能各方面知识和技术，培养相关人才。训练营联合开源项目开发者、学术界大牛、工业界资深工程师参与授课与项目指导。课程采用「线上直播授课 + 在线答疑 + 项目实践辅导 」的方式进行，确保每位同学都能获得充分成长。训练营全程免费，通过视频、直播课、习题和项目实践，在人工智能飞速发展的背景下，为学员提供一窥深层技术的机会。","北京积算科技有限公司","2025-07-14 08:00:00","2025-09-30 08:00:00","upcoming","2025-07-13 08:00:00","2026-10-22 20:24:58",{"id":85,"slug":86,"title":87,"logo":88,"publicLiveEmbedUrl":7,"description":89,"introduction":57,"organizer":18,"coOrganizer":5,"sponsor":5,"startTime":90,"endTime":91,"status":81,"dashboardVisibility":23,"createdAt":92,"updatedAt":93},"01980cf9-a7d8-7796-ae86-0bbe05334add","winter2024","2024年冬季InfiniTensor大模型与人工智能系统训练营","https:\u002F\u002Ftesting-cdn.infinitensor.com\u002Fuploads\u002F2025-07\u002F01981269-ce23-7624-a287-9f07dd0d85f6.jpg","InfiniTensor大模型与人工智能系统训练营由清华大学发起，面向开源社区、高校和产业界，普及人工智能各方面知识和技术，培养相关人才。训练营全程免费，通过视频、直播课、习题和项目实践，在人工智能飞速发展的背景下，为学员提供一窥深层技术的机会。报名学员请阅读“导学”课程页的训练营介绍。","2025-01-05 08:00:00","2025-03-22 08:00:00","2025-07-15 08:00:00","2026-10-22 20:24:39",{"id":95,"slug":96,"title":97,"logo":98,"publicLiveEmbedUrl":7,"description":99,"introduction":57,"organizer":18,"coOrganizer":5,"sponsor":5,"startTime":100,"endTime":101,"status":81,"dashboardVisibility":23,"createdAt":92,"updatedAt":102},"01980c20-363a-7dc3-8f5a-58a68e08dcc4","summer2024","2024年夏季InfiniTensor大模型与人工智能系统训练营","https:\u002F\u002Ftesting-cdn.infinitensor.com\u002Fuploads\u002F2025-07\u002F01981269-fbfa-788a-afcd-17fa9be74ff9.jpg","InfiniTensor大模型与人工智能系统训练营由清华大学发起，面向开源社区和高校，普及人工智能各方面知识和技术，培养相关人才。训练营全程免费，通过课程、习题和项目实践，在人工智能飞速发展的背景下，为学员提供一窥深层技术的机会。","2024-07-20 08:00:00","2024-09-29 08:00:00","2026-10-22 20:24:49",[44],{"currentItem":105,"processedData":125},{"img":106,"keywords":107,"treeType":44,"description":108,"showStatus":44,"updateTime":109,"type":110,"title":111,"content":112,"nextArticle":113,"createTime":109,"sortOrder":117,"canonicalPath":118,"id":119,"contentType":120,"prevArticle":121},"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_fm.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=VPQUx3mNrL7p08nFbuVCG8goOK8%3D","CPU 性能优化,SIMD,矩阵分块,模型量化,多线程并行","面向智能体时代讲解 CPU 全套优化方案，覆盖多级缓存、局部性、SIMD、多线程、矩阵分块、PTQ\u002FQAT 量化、NUMA 架构，配套工程落地优化思路。","2026-10-22 20:38:11",2,"2026 夏季 InfiniTensor 训练营 CPU 并行编程 1：《CPU 性能优化导论》","![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw01.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=JbJLwOAh01lETaekWVN9KGPRfXI%3D)\r\n\r\n\u003Cspan style=\"font-size:12px;color:#555;\">7 月 17 日，2026 夏季 InfiniTensor 训练营 CPU 并行编程方向第一课《CPU 性能优化导论》开讲。\r\n本课程将带你从底层原理出发，完整拆解 CPU 多级缓存、局部性原理、SIMD 向量指令、多线程并行、矩阵分块、模型量化六大核心优化手段，搭配真实工程案例讲解，掌握 Agent 时代的 CPU 性能优化方法论。\u003C\u002Fspan>\r\n\r\n\r\n## Agent 时代的计算新范式\r\n#### 1.从 Chatbot 到 Agent 的负载变迁\r\n- 传统 Chatbot：单次请求 - 响应模式，侧重模型推理。\r\n- Agent 模式：工具调用 + 本地\u002F云侧数据处理 + 循环执行，侧重全链路编排。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw02.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Onn%2BMOov%2BvHs34lyUUSqDX%2Bh9rQ%3D)\r\n\r\n#### 2.为什么 CPU 依然是核心？\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw03.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=x5KmpUYfqu5ERze%2BJfpjufyOS%2Fs%3D)\r\n\r\n- 复杂计算负载的调度中枢（浏览器操作、文件访问、API 调用）。\r\n- 长期在线与完整工具链运行的算力基石。\r\n- 传统架构痛点：CPU\u002FGPU 内存通常不互通，数据搬运开销大。\r\n- 统一内存优势：CPU\u002FGPU\u002FNPU 共享内存池，上下文切换效率高（以 Mac mini 为例）。\r\n- 潜在挑战：多计算单元共享带来的资源竞争问题。\r\n\r\n#### 3. CPU 性能优化的核心维度\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw04.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=%2Bao4vlDm93wwullhJwGGGPq%2BZ%2F4%3D)\r\n\r\n-  **时间维度的拆解**\r\n   -  计算时间：CPU 负载是否有效拉满？避免无效空转。\r\n   - 等待时间：减少内存等待与 I\u002FO 等待（IO Bound 问题）。\r\n   - 异步处理策略：在等待网络\u002FIO 时合理调度其他任务。\r\n\r\n- **硬件性能的充分释放**\r\n  - 寄存器、Cache 与内存的数据交换效率。\r\n  - 端侧 AI 设备的多传感器与计算单元协同。\r\n  - 核心理念：理解基础原理是写出高效代码的前提。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw05.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=jQ0wNZHMUu7%2Bxvh1OZ8ef4K8r9s%3D)\r\n\r\n## 存储层级与局部性原理\r\n### CPU 多级缓存结构回顾\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw06.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=6DpiJqt26JGYs38a0dxbUgIb5fA%3D)\r\n\r\n1. 存储金字塔：寄存器 → L1\u002FL2\u002FL3 Cache → 内存 → 硬盘。\r\n2. 访问延迟差异与“备菜理论”（数据预取的重要性）。\r\n\r\n### 局部性原则（Locality Principle）\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw07.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=PN6JV2%2BVQ7GhDPe7k4vB2yt%2BGhs%3D)\r\n\r\n1. **时间局部性**：刚被访问的数据很可能再次被访问。\r\n2. **空间局部性**：连续内存访问效率远高于跨步\u002F随机访问。\r\n3. **目标**：最大化 Cache Hit，最小化 Cache Miss。\r\n4. **二维矩阵存储布局**：\r\n   - 行主序（Row-major）vs 列主序（Column-major）的区别。\r\n   - 循环嵌套顺序对 Cache 命中的影响。\r\n   - 调用高性能计算库（BLAS）时的内存排布注意事项。\r\n\r\n\r\n## 指令集加速与并行计算\r\n### SIMD 向量指令集\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw08.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Q%2BYJ8cqwuLrK9RZjx23G2XCIg%2BY%3D)\r\n\r\n1. Single Instruction Multi Data 原理：一条指令处理多个数据。\r\n2. 位宽与吞吐量：128bit\u002F256bit 指令集具有理论性能倍增效果。\r\n3. FMA（融合乘加）指令在矩阵运算中的应用。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw09.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Vfj3O7S5ROlLJvqyUp4%2FMOo95VE%3D)\r\n\r\n### 多线程并行技术\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw10.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=m%2B5dA2NdIQwDq6uEYbRjzwHLLEY%3D)\r\n\r\n1. 线程池机制：避免频繁创建\u002F销毁线程的开销\r\n2. 任务分配策略：独立任务的并行执行与负载均衡。\r\n3. 线程同步与并发安全\r\n   - 临界区与数据竞争问题。\r\n   - 同步机制：互斥锁（Mutex）、信号量、Barrier。\r\n   - 算子融合与大模型推理中的同步设计模式。\r\n\r\n## 矩阵运算深度优化\r\n### 分块（Blocking\u002FTiling）技术\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw11.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=x1aDwc7sWCxHkiEwe9ufiGpLoxY%3D)\r\n\r\n1. 将矩阵切分为适配 L1\u002FL2 Cache 的小块。\r\n2. 减少数据搬运，提升计算单元利用率。\r\n3. 内外循环设计与分块矩阵乘法理论验证。\r\n\r\n### 数据量化与精度优化\r\n#### 1. 量化基础\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw12.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=whIJYFVFZiD4PZmIXC37p%2FT0SS0%3D)\r\n\r\n- 高精度计算向低精度表示转换，包括 FP16、BF16 等低精度浮点计算，以及 INT8、INT4 等模型量化形式。\r\n- 收益：体积缩减 + 计算周期缩短。\r\n- 容错率：神经网络参数和激活通常存在一定数值冗余，模型对有限的舍入误差具有一定容忍能力，可通过校准或量化感知训练降低精度损失。\r\n\r\n#### 2. 主流量化方法\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw13.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=LARu6vk%2BCUtVk5kc4S%2BwbsYToJQ%3D)\r\n\r\n- **PTQ（训练后量化）**：校准数据集 + 权重映射。\r\n- **QAT（量化感知训练）**：在训练过程加入了模拟量化。\r\n- **常见量化组合（W x A x）**\r\n  - W4A16 \u002F W8A16 \u002F W8A8 等权重-激活值组合。\r\n  - 硬件指令集对量化运算的支持情况。\r\n\r\n#### 3.其他优化方法\r\n- **内存对齐（Alignment）**\r\n  - 非对齐访问导致的 Cache 抖动与性能损耗。\r\n  - 强制对齐指令与编译器优化。\r\n\r\n-  **非一致性内存访问 NUMA**\r\n   - 多路服务器跨节点访问的高延迟问题。\r\n   - 高速互联技术下的数据编排策略。\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw14.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=YrhOf%2BRcKjiQ6bXrnEDHwCKeaZE%3D)\r\n\r\n\r\n## 开源框架\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw15.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=CLiR5sZ2AByGbWztJw3L3YOTlv0%3D)\r\n\r\n\r\n\r\n\r\n## 加入训练营\r\n#### 📺完整课程内容，请观看直播或查看课程回放：\r\n**InfiniTensor 官网**： https:\u002F\u002Fwww.infinitensor.com\r\n\r\n**B站直播**： InfiniTensor 官方直播间\r\n\r\n**视频号直播**： InfiniTensor 视频号预约\r\n\r\n**答疑交流**： 训练营官方社群⬇️\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002Fportal_news\u002F2026-xjq.png?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=oKlo6p8Ra0fcURk4F6PNTseTp%2BA%3D)\r\n\r\n## 关注与交流\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002Fportal_news\u002Fnews_ewm02.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=R03Psa4VwXKPuVbshfyqCCO33b4%3D)",{"publishTime":7,"createTime":109,"canonicalPath":114,"id":115,"title":116},"\u002Fstudy-material\u002Fstudy-65","study-65","2026 夏季 InfiniTensor 训练营 Triton & 九齿 1：《Triton 练气术》",43,"\u002Fstudy-material\u002Fstudy-66","study-66","markdown",{"publishTime":7,"createTime":109,"canonicalPath":122,"id":123,"title":124},"\u002Fstudy-material\u002Fstudy-67","study-67","2026 夏季 InfiniTensor 训练营前沿模型 1：《大模型学习方法》",{"htmlContent":126,"headings":127},"\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw01.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=JbJLwOAh01lETaekWVN9KGPRfXI%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>\u003Cspan style=\"font-size:12px;color:#555;\">7 月 17 日，2026 夏季 InfiniTensor 训练营 CPU 并行编程方向第一课《CPU 性能优化导论》开讲。\u003Cbr>本课程将带你从底层原理出发，完整拆解 CPU 多级缓存、局部性原理、SIMD 向量指令、多线程并行、矩阵分块、模型量化六大核心优化手段，搭配真实工程案例讲解，掌握 Agent 时代的 CPU 性能优化方法论。\u003C\u002Fspan>\u003C\u002Fp>\n\u003Ch2 id=\"agent-时代的计算新范式\">\u003Ca class=\"anchor\" href=\"#agent-时代的计算新范式\" aria-label=\"跳转到标题：Agent 时代的计算新范式\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Agent 时代的计算新范式\u003C\u002Fh2>\u003Ch4 id=\"1-从-chatbot-到-agent-的负载变迁\">\u003Ca class=\"anchor\" href=\"#1-从-chatbot-到-agent-的负载变迁\" aria-label=\"跳转到标题：1.从 Chatbot 到 Agent 的负载变迁\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1.从 Chatbot 到 Agent 的负载变迁\u003C\u002Fh4>\u003Cul>\n\u003Cli>传统 Chatbot：单次请求 - 响应模式，侧重模型推理。\u003C\u002Fli>\n\u003Cli>Agent 模式：工具调用 + 本地\u002F云侧数据处理 + 循环执行，侧重全链路编排。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw02.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Onn%2BMOov%2BvHs34lyUUSqDX%2Bh9rQ%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"2-为什么-cpu-依然是核心\">\u003Ca class=\"anchor\" href=\"#2-为什么-cpu-依然是核心\" aria-label=\"跳转到标题：2.为什么 CPU 依然是核心？\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2.为什么 CPU 依然是核心？\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw03.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=x5KmpUYfqu5ERze%2BJfpjufyOS%2Fs%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>复杂计算负载的调度中枢（浏览器操作、文件访问、API 调用）。\u003C\u002Fli>\n\u003Cli>长期在线与完整工具链运行的算力基石。\u003C\u002Fli>\n\u003Cli>传统架构痛点：CPU\u002FGPU 内存通常不互通，数据搬运开销大。\u003C\u002Fli>\n\u003Cli>统一内存优势：CPU\u002FGPU\u002FNPU 共享内存池，上下文切换效率高（以 Mac mini 为例）。\u003C\u002Fli>\n\u003Cli>潜在挑战：多计算单元共享带来的资源竞争问题。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch4 id=\"3-cpu-性能优化的核心维度\">\u003Ca class=\"anchor\" href=\"#3-cpu-性能优化的核心维度\" aria-label=\"跳转到标题：3. CPU 性能优化的核心维度\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>3. CPU 性能优化的核心维度\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw04.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=%2Bao4vlDm93wwullhJwGGGPq%2BZ%2F4%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cp>\u003Cstrong>时间维度的拆解\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cul>\n\u003Cli>计算时间：CPU 负载是否有效拉满？避免无效空转。\u003C\u002Fli>\n\u003Cli>等待时间：减少内存等待与 I\u002FO 等待（IO Bound 问题）。\u003C\u002Fli>\n\u003Cli>异步处理策略：在等待网络\u002FIO 时合理调度其他任务。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003C\u002Fli>\n\u003Cli>\u003Cp>\u003Cstrong>硬件性能的充分释放\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cul>\n\u003Cli>寄存器、Cache 与内存的数据交换效率。\u003C\u002Fli>\n\u003Cli>端侧 AI 设备的多传感器与计算单元协同。\u003C\u002Fli>\n\u003Cli>核心理念：理解基础原理是写出高效代码的前提。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw05.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=jQ0wNZHMUu7%2Bxvh1OZ8ef4K8r9s%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 id=\"存储层级与局部性原理\">\u003Ca class=\"anchor\" href=\"#存储层级与局部性原理\" aria-label=\"跳转到标题：存储层级与局部性原理\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>存储层级与局部性原理\u003C\u002Fh2>\u003Ch3 id=\"cpu-多级缓存结构回顾\">\u003Ca class=\"anchor\" href=\"#cpu-多级缓存结构回顾\" aria-label=\"跳转到标题：CPU 多级缓存结构回顾\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>CPU 多级缓存结构回顾\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw06.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=6DpiJqt26JGYs38a0dxbUgIb5fA%3D\" alt=\"\">\u003C\u002Fp>\n\u003Col>\n\u003Cli>存储金字塔：寄存器 → L1\u002FL2\u002FL3 Cache → 内存 → 硬盘。\u003C\u002Fli>\n\u003Cli>访问延迟差异与“备菜理论”（数据预取的重要性）。\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Ch3 id=\"局部性原则-locality-principle\">\u003Ca class=\"anchor\" href=\"#局部性原则-locality-principle\" aria-label=\"跳转到标题：局部性原则（Locality Principle）\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>局部性原则（Locality Principle）\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw07.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=PN6JV2%2BVQ7GhDPe7k4vB2yt%2BGhs%3D\" alt=\"\">\u003C\u002Fp>\n\u003Col>\n\u003Cli>\u003Cstrong>时间局部性\u003C\u002Fstrong>：刚被访问的数据很可能再次被访问。\u003C\u002Fli>\n\u003Cli>\u003Cstrong>空间局部性\u003C\u002Fstrong>：连续内存访问效率远高于跨步\u002F随机访问。\u003C\u002Fli>\n\u003Cli>\u003Cstrong>目标\u003C\u002Fstrong>：最大化 Cache Hit，最小化 Cache Miss。\u003C\u002Fli>\n\u003Cli>\u003Cstrong>二维矩阵存储布局\u003C\u002Fstrong>：\u003Cul>\n\u003Cli>行主序（Row-major）vs 列主序（Column-major）的区别。\u003C\u002Fli>\n\u003Cli>循环嵌套顺序对 Cache 命中的影响。\u003C\u002Fli>\n\u003Cli>调用高性能计算库（BLAS）时的内存排布注意事项。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Ch2 id=\"指令集加速与并行计算\">\u003Ca class=\"anchor\" href=\"#指令集加速与并行计算\" aria-label=\"跳转到标题：指令集加速与并行计算\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>指令集加速与并行计算\u003C\u002Fh2>\u003Ch3 id=\"simd-向量指令集\">\u003Ca class=\"anchor\" href=\"#simd-向量指令集\" aria-label=\"跳转到标题：SIMD 向量指令集\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>SIMD 向量指令集\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw08.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Q%2BYJ8cqwuLrK9RZjx23G2XCIg%2BY%3D\" alt=\"\">\u003C\u002Fp>\n\u003Col>\n\u003Cli>Single Instruction Multi Data 原理：一条指令处理多个数据。\u003C\u002Fli>\n\u003Cli>位宽与吞吐量：128bit\u002F256bit 指令集具有理论性能倍增效果。\u003C\u002Fli>\n\u003Cli>FMA（融合乘加）指令在矩阵运算中的应用。\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw09.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Vfj3O7S5ROlLJvqyUp4%2FMOo95VE%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"多线程并行技术\">\u003Ca class=\"anchor\" href=\"#多线程并行技术\" aria-label=\"跳转到标题：多线程并行技术\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>多线程并行技术\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw10.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=m%2B5dA2NdIQwDq6uEYbRjzwHLLEY%3D\" alt=\"\">\u003C\u002Fp>\n\u003Col>\n\u003Cli>线程池机制：避免频繁创建\u002F销毁线程的开销\u003C\u002Fli>\n\u003Cli>任务分配策略：独立任务的并行执行与负载均衡。\u003C\u002Fli>\n\u003Cli>线程同步与并发安全\u003Cul>\n\u003Cli>临界区与数据竞争问题。\u003C\u002Fli>\n\u003Cli>同步机制：互斥锁（Mutex）、信号量、Barrier。\u003C\u002Fli>\n\u003Cli>算子融合与大模型推理中的同步设计模式。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Ch2 id=\"矩阵运算深度优化\">\u003Ca class=\"anchor\" href=\"#矩阵运算深度优化\" aria-label=\"跳转到标题：矩阵运算深度优化\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>矩阵运算深度优化\u003C\u002Fh2>\u003Ch3 id=\"分块-blocking-tiling-技术\">\u003Ca class=\"anchor\" href=\"#分块-blocking-tiling-技术\" aria-label=\"跳转到标题：分块（Blocking\u002FTiling）技术\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>分块（Blocking\u002FTiling）技术\u003C\u002Fh3>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw11.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=x1aDwc7sWCxHkiEwe9ufiGpLoxY%3D\" alt=\"\">\u003C\u002Fp>\n\u003Col>\n\u003Cli>将矩阵切分为适配 L1\u002FL2 Cache 的小块。\u003C\u002Fli>\n\u003Cli>减少数据搬运，提升计算单元利用率。\u003C\u002Fli>\n\u003Cli>内外循环设计与分块矩阵乘法理论验证。\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Ch3 id=\"数据量化与精度优化\">\u003Ca class=\"anchor\" href=\"#数据量化与精度优化\" aria-label=\"跳转到标题：数据量化与精度优化\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>数据量化与精度优化\u003C\u002Fh3>\u003Ch4 id=\"1-量化基础\">\u003Ca class=\"anchor\" href=\"#1-量化基础\" aria-label=\"跳转到标题：1. 量化基础\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1. 量化基础\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw12.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=whIJYFVFZiD4PZmIXC37p%2FT0SS0%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>高精度计算向低精度表示转换，包括 FP16、BF16 等低精度浮点计算，以及 INT8、INT4 等模型量化形式。\u003C\u002Fli>\n\u003Cli>收益：体积缩减 + 计算周期缩短。\u003C\u002Fli>\n\u003Cli>容错率：神经网络参数和激活通常存在一定数值冗余，模型对有限的舍入误差具有一定容忍能力，可通过校准或量化感知训练降低精度损失。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch4 id=\"2-主流量化方法\">\u003Ca class=\"anchor\" href=\"#2-主流量化方法\" aria-label=\"跳转到标题：2. 主流量化方法\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2. 主流量化方法\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw13.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=LARu6vk%2BCUtVk5kc4S%2BwbsYToJQ%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>PTQ（训练后量化）\u003C\u002Fstrong>：校准数据集 + 权重映射。\u003C\u002Fli>\n\u003Cli>\u003Cstrong>QAT（量化感知训练）\u003C\u002Fstrong>：在训练过程加入了模拟量化。\u003C\u002Fli>\n\u003Cli>\u003Cstrong>常见量化组合（W x A x）\u003C\u002Fstrong>\u003Cul>\n\u003Cli>W4A16 \u002F W8A16 \u002F W8A8 等权重-激活值组合。\u003C\u002Fli>\n\u003Cli>硬件指令集对量化运算的支持情况。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch4 id=\"3-其他优化方法\">\u003Ca class=\"anchor\" href=\"#3-其他优化方法\" aria-label=\"跳转到标题：3.其他优化方法\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>3.其他优化方法\u003C\u002Fh4>\u003Cul>\n\u003Cli>\u003Cp>\u003Cstrong>内存对齐（Alignment）\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cul>\n\u003Cli>非对齐访问导致的 Cache 抖动与性能损耗。\u003C\u002Fli>\n\u003Cli>强制对齐指令与编译器优化。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003C\u002Fli>\n\u003Cli>\u003Cp>\u003Cstrong>非一致性内存访问 NUMA\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cul>\n\u003Cli>多路服务器跨节点访问的高延迟问题。\u003C\u002Fli>\n\u003Cli>高速互联技术下的数据编排策略。\u003C\u002Fli>\n\u003C\u002Ful>\n\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw14.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=YrhOf%2BRcKjiQ6bXrnEDHwCKeaZE%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 id=\"开源框架\">\u003Ca class=\"anchor\" href=\"#开源框架\" aria-label=\"跳转到标题：开源框架\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>开源框架\u003C\u002Fh2>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_66_zw15.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=CLiR5sZ2AByGbWztJw3L3YOTlv0%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 id=\"加入训练营\">\u003Ca class=\"anchor\" href=\"#加入训练营\" aria-label=\"跳转到标题：加入训练营\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>加入训练营\u003C\u002Fh2>\u003Ch4 id=\"完整课程内容-请观看直播或查看课程回放\">\u003Ca class=\"anchor\" href=\"#完整课程内容-请观看直播或查看课程回放\" aria-label=\"跳转到标题：📺完整课程内容，请观看直播或查看课程回放：\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>📺完整课程内容，请观看直播或查看课程回放：\u003C\u002Fh4>\u003Cp>\u003Cstrong>InfiniTensor 官网\u003C\u002Fstrong>： \u003Ca href=\"https:\u002F\u002Fwww.infinitensor.com\">https:\u002F\u002Fwww.infinitensor.com\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>\u003Cstrong>B站直播\u003C\u002Fstrong>： InfiniTensor 官方直播间\u003C\u002Fp>\n\u003Cp>\u003Cstrong>视频号直播\u003C\u002Fstrong>： InfiniTensor 视频号预约\u003C\u002Fp>\n\u003Cp>\u003Cstrong>答疑交流\u003C\u002Fstrong>： 训练营官方社群⬇️\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002Fportal_news\u002F2026-xjq.png?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=oKlo6p8Ra0fcURk4F6PNTseTp%2BA%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch2 id=\"关注与交流\">\u003Ca class=\"anchor\" href=\"#关注与交流\" aria-label=\"跳转到标题：关注与交流\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>关注与交流\u003C\u002Fh2>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002Fportal_news\u002Fnews_ewm02.jpg?Expires=1792752557&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=R03Psa4VwXKPuVbshfyqCCO33b4%3D\" alt=\"\">\u003C\u002Fp>\n",[128,131,135,138,141,143,146,149,151,154,156,158,161,163,166,169,172,174,176,179],{"id":129,"text":130,"level":110},"agent-时代的计算新范式","Agent 时代的计算新范式",{"id":132,"text":133,"level":134},"1-从-chatbot-到-agent-的负载变迁","1.从 Chatbot 到 Agent 的负载变迁",4,{"id":136,"text":137,"level":134},"2-为什么-cpu-依然是核心","2.为什么 CPU 依然是核心？",{"id":139,"text":140,"level":134},"3-cpu-性能优化的核心维度","3. CPU 性能优化的核心维度",{"id":142,"text":142,"level":110},"存储层级与局部性原理",{"id":144,"text":145,"level":9},"cpu-多级缓存结构回顾","CPU 多级缓存结构回顾",{"id":147,"text":148,"level":9},"局部性原则-locality-principle","局部性原则（Locality Principle）",{"id":150,"text":150,"level":110},"指令集加速与并行计算",{"id":152,"text":153,"level":9},"simd-向量指令集","SIMD 向量指令集",{"id":155,"text":155,"level":9},"多线程并行技术",{"id":157,"text":157,"level":110},"矩阵运算深度优化",{"id":159,"text":160,"level":9},"分块-blocking-tiling-技术","分块（Blocking\u002FTiling）技术",{"id":162,"text":162,"level":9},"数据量化与精度优化",{"id":164,"text":165,"level":134},"1-量化基础","1. 量化基础",{"id":167,"text":168,"level":134},"2-主流量化方法","2. 主流量化方法",{"id":170,"text":171,"level":134},"3-其他优化方法","3.其他优化方法",{"id":173,"text":173,"level":110},"开源框架",{"id":175,"text":175,"level":110},"加入训练营",{"id":177,"text":178,"level":134},"完整课程内容-请观看直播或查看课程回放","📺完整课程内容，请观看直播或查看课程回放：",{"id":180,"text":180,"level":110},"关注与交流"]