[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"site-home-live-entry-setting":3,"site-header-contests":8,"site-header-camps":49,"study-material-detail-study-17":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=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=YuOgm%2FgmD4nuySjwNMl5nm0fmKI%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=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Kxv2UTqtU0%2FwvYQCDERqdg0W6E4%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=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=nIij3MTYWjwdhBM00gMTqmyLnUM%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_17_fm.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=cKoXWv%2Fn3leekMEPBr3jHK4ABUU%3D","InfiniTensor,冬季训练营,CUDA,性能模型,逐元素优化","围绕 CUDA 性能模型与逐元素优化展开，讲解 kernel 设计与热点优化实践。","2026-10-22 20:38:11",2,"InfiniTensor 冬季训练营 CUDA 编程第二课《性能模型与逐元素优化》","![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw01.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=pgGINJoJ7zp6C2jqdP4os6Cz6Hc%3D)\r\n\r\n2026 年 1 月 15 日，2025 冬季 InfiniTensor 大模型与人工智能系统训练营 CUDA 编程系列第二课《性能模型与逐元素优化》开讲。\r\n\r\n本节课涵盖并行编程基础、Roofline 模型等内容，并涉及向量化实现与性能对比等知识。\r\n\r\n### 课程回顾与背景引入\r\n- 什么是并行编程？为什么需要并行？\r\n- CPU 与 GPU 的架构差异\r\n- 向量加法的 GPU 实现与编译\r\n- 性能初步分析工具\r\n- CPU 与 GPU 性能对比\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw02.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=amNPBfwR6lWU7R3vGNc104rxn1s%3D)\r\n\r\n### 性能瓶颈分析与内存墙\r\n#### 1. 传输开销与计算时间对比\r\n- CPU-GPU 数据传输时间 vs  GPU 内核计算时间\r\n- 通信带宽成为性能瓶颈\r\n#### 2. 内存墙概念引入\r\n- 摩尔定律与内存增长速度对比\r\n- 冯·诺依曼架构的限制\r\n- 计算速度远快于内存访问速度\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw03.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=a7RaLTXmtcPID61i6a2AD8esLi4%3D)\r\n\r\n### Roofline 模型：量化性能瓶颈\r\n#### 1. Roofline 模型基本概念\r\n- 纵轴：性能（GFLOPS）\r\n- 横轴：计算强度（FLOPS\u002Fbyte）\r\n- 关键参数：\r\n  - π（峰值算力）\r\n  - β（峰值内存带宽）\r\n  - Imax（计算强度拐点）\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw04.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=TSEUZKeerBcVIxWh8b%2Fez3GsYX8%3D)\r\n\r\n#### 2. 模型分区解释\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw05.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=d2WE05bMIsGG3ESSPyhlr8de2Vs%3D)\r\n\r\n#### 3. 向量加法案例分析\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw06.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=yoYJ7R%2Bar0eBesaMcvOZ9cuAEk4%3D)\r\n\r\n\r\n### 性能分析工具：Nsight Compute（NCU）\r\n#### 1. NCU 简介与使用\r\n- 内核级性能分析工具\r\n- 命令行与 GUI 版本使用方式\r\n- 输出文件分析\r\n#### 2. 关键分析界面\r\n- Summary Page（核函数汇总）\r\n- Details Page（吞吐量与 Roofline 图）\r\n- 内存与计算利用率分析\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw07.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=vwnp0OwkmVVgcnw6fQXPV6lCmps%3D)\r\n\r\n### 向量化技术：提升访存效率\r\n#### 1. 什么是向量化？\r\n- 标量操作 vs 向量操作\r\n- SIMD（单指令多数据） vs  SIMT（单指令多线程）\r\n#### 2. CUDA 中的向量化访存类型\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw08.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=KB9OPz3t64E65pRDUwPhKHoWh38%3D)\r\n\r\n#### 3. 向量化实现与性能对比\r\n- 不同向量化程度的性能测试\r\n- 寄存器使用与溢出问题\r\n- 数据规模对向量化效果的影响\r\n\r\n### 半精度计算\r\n#### 1. 半精度浮点数简介\r\n- 结构对比（FP16 vs FP32）\r\n- CUDA 中的 half 类型与转换函数\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw09.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=OPnjbsp2zMEcW74nq0LzYJfMRKQ%3D)\r\n\r\n#### 2. 半精度性能优势\r\n- 内存占用减半\r\n- 计算强度提升\r\n- 向量化支持（half2）\r\n#### 3. 向量化计算与访存结合\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw10.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=c8YD77sJCsrU3kpMpcTv%2FrG48L4%3D)\r\n\r\n### 课后思考题\r\n1.可以有效使用 float3 吗?如果可以，怎么做?\r\n2.之前提到向量化访存有多种方法，今天主要介绍了使用内置类型的方式，想想还有什么其他方式吗?(Hint: 今天的课中其实已经提到了，同时也和第一问有关)\r\n\r\n### 总结与思考\r\n本节课我们对优化路径进行了系统性学习，向量化访存 → 半精度 → 向量化计算，性能提升幅度对比等知识。后续课程我们将进一步对于优化方法，算子融合、内存层级优化等知识进行学习，同时将会迎来新的挑战！\r\n\r\n\r\n![2025冬季训练营课程表](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_01_zw6.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Ffj54EQtw2SzZEJdvnMd%2BBMhib0%3D)\r\n\r\n### 💡 学习贴士\r\n更多细节及完整课程内容，请观看直播或查看课程回放。\r\n\r\n链接：https:\u002F\u002Fwww.infinitensor.com\u002Fcamp\u002Fwinter2025\r\n\r\n### 📚 训练营报名持续进行中\r\n扫码报名与加入群聊，或点击 **「阅读原文」** 直达官网报名\r\n\r\n\r\n![](https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_01_zw7.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=ZU5%2F7VUywJvItSFp2MjcU%2FW%2FeuQ%3D)",{"publishTime":7,"createTime":109,"canonicalPath":114,"id":115,"title":116},"\u002Fstudy-material\u002Fstudy-16","study-16","InfiniTensor 冬季训练营前沿模型第二课《科学多模态大模型》",92,"\u002Fstudy-material\u002Fstudy-17","study-17","markdown",{"publishTime":7,"createTime":109,"canonicalPath":122,"id":123,"title":124},"\u002Fstudy-material\u002Fstudy-18","study-18","InfiniTensor 冬季训练营 AI 编译器第二课《AI 编译器中的前端优化》",{"htmlContent":126,"headings":127},"\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw01.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=pgGINJoJ7zp6C2jqdP4os6Cz6Hc%3D\" alt=\"\">\u003C\u002Fp>\n\u003Cp>2026 年 1 月 15 日，2025 冬季 InfiniTensor 大模型与人工智能系统训练营 CUDA 编程系列第二课《性能模型与逐元素优化》开讲。\u003C\u002Fp>\n\u003Cp>本节课涵盖并行编程基础、Roofline 模型等内容，并涉及向量化实现与性能对比等知识。\u003C\u002Fp>\n\u003Ch3 id=\"课程回顾与背景引入\">\u003Ca class=\"anchor\" href=\"#课程回顾与背景引入\" aria-label=\"跳转到标题：课程回顾与背景引入\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>课程回顾与背景引入\u003C\u002Fh3>\u003Cul>\n\u003Cli>什么是并行编程？为什么需要并行？\u003C\u002Fli>\n\u003Cli>CPU 与 GPU 的架构差异\u003C\u002Fli>\n\u003Cli>向量加法的 GPU 实现与编译\u003C\u002Fli>\n\u003Cli>性能初步分析工具\u003C\u002Fli>\n\u003Cli>CPU 与 GPU 性能对比\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw02.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=amNPBfwR6lWU7R3vGNc104rxn1s%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"性能瓶颈分析与内存墙\">\u003Ca class=\"anchor\" href=\"#性能瓶颈分析与内存墙\" aria-label=\"跳转到标题：性能瓶颈分析与内存墙\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>性能瓶颈分析与内存墙\u003C\u002Fh3>\u003Ch4 id=\"1-传输开销与计算时间对比\">\u003Ca class=\"anchor\" href=\"#1-传输开销与计算时间对比\" aria-label=\"跳转到标题：1. 传输开销与计算时间对比\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1. 传输开销与计算时间对比\u003C\u002Fh4>\u003Cul>\n\u003Cli>CPU-GPU 数据传输时间 vs  GPU 内核计算时间\u003C\u002Fli>\n\u003Cli>通信带宽成为性能瓶颈\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch4 id=\"2-内存墙概念引入\">\u003Ca class=\"anchor\" href=\"#2-内存墙概念引入\" aria-label=\"跳转到标题：2. 内存墙概念引入\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2. 内存墙概念引入\u003C\u002Fh4>\u003Cul>\n\u003Cli>摩尔定律与内存增长速度对比\u003C\u002Fli>\n\u003Cli>冯·诺依曼架构的限制\u003C\u002Fli>\n\u003Cli>计算速度远快于内存访问速度\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw03.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=a7RaLTXmtcPID61i6a2AD8esLi4%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"roofline-模型-量化性能瓶颈\">\u003Ca class=\"anchor\" href=\"#roofline-模型-量化性能瓶颈\" aria-label=\"跳转到标题：Roofline 模型：量化性能瓶颈\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>Roofline 模型：量化性能瓶颈\u003C\u002Fh3>\u003Ch4 id=\"1-roofline-模型基本概念\">\u003Ca class=\"anchor\" href=\"#1-roofline-模型基本概念\" aria-label=\"跳转到标题：1. Roofline 模型基本概念\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1. Roofline 模型基本概念\u003C\u002Fh4>\u003Cul>\n\u003Cli>纵轴：性能（GFLOPS）\u003C\u002Fli>\n\u003Cli>横轴：计算强度（FLOPS\u002Fbyte）\u003C\u002Fli>\n\u003Cli>关键参数：\u003Cul>\n\u003Cli>π（峰值算力）\u003C\u002Fli>\n\u003Cli>β（峰值内存带宽）\u003C\u002Fli>\n\u003Cli>Imax（计算强度拐点）\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_17_zw04.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=TSEUZKeerBcVIxWh8b%2Fez3GsYX8%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"2-模型分区解释\">\u003Ca class=\"anchor\" href=\"#2-模型分区解释\" aria-label=\"跳转到标题：2. 模型分区解释\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2. 模型分区解释\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw05.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=d2WE05bMIsGG3ESSPyhlr8de2Vs%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"3-向量加法案例分析\">\u003Ca class=\"anchor\" href=\"#3-向量加法案例分析\" aria-label=\"跳转到标题：3. 向量加法案例分析\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>3. 向量加法案例分析\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw06.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=yoYJ7R%2Bar0eBesaMcvOZ9cuAEk4%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"性能分析工具-nsight-compute-ncu\">\u003Ca class=\"anchor\" href=\"#性能分析工具-nsight-compute-ncu\" aria-label=\"跳转到标题：性能分析工具：Nsight Compute（NCU）\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>性能分析工具：Nsight Compute（NCU）\u003C\u002Fh3>\u003Ch4 id=\"1-ncu-简介与使用\">\u003Ca class=\"anchor\" href=\"#1-ncu-简介与使用\" aria-label=\"跳转到标题：1. NCU 简介与使用\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1. NCU 简介与使用\u003C\u002Fh4>\u003Cul>\n\u003Cli>内核级性能分析工具\u003C\u002Fli>\n\u003Cli>命令行与 GUI 版本使用方式\u003C\u002Fli>\n\u003Cli>输出文件分析\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch4 id=\"2-关键分析界面\">\u003Ca class=\"anchor\" href=\"#2-关键分析界面\" aria-label=\"跳转到标题：2. 关键分析界面\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2. 关键分析界面\u003C\u002Fh4>\u003Cul>\n\u003Cli>Summary Page（核函数汇总）\u003C\u002Fli>\n\u003Cli>Details Page（吞吐量与 Roofline 图）\u003C\u002Fli>\n\u003Cli>内存与计算利用率分析\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw07.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=vwnp0OwkmVVgcnw6fQXPV6lCmps%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"向量化技术-提升访存效率\">\u003Ca class=\"anchor\" href=\"#向量化技术-提升访存效率\" aria-label=\"跳转到标题：向量化技术：提升访存效率\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>向量化技术：提升访存效率\u003C\u002Fh3>\u003Ch4 id=\"1-什么是向量化\">\u003Ca class=\"anchor\" href=\"#1-什么是向量化\" aria-label=\"跳转到标题：1. 什么是向量化？\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1. 什么是向量化？\u003C\u002Fh4>\u003Cul>\n\u003Cli>标量操作 vs 向量操作\u003C\u002Fli>\n\u003Cli>SIMD（单指令多数据） vs  SIMT（单指令多线程）\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch4 id=\"2-cuda-中的向量化访存类型\">\u003Ca class=\"anchor\" href=\"#2-cuda-中的向量化访存类型\" aria-label=\"跳转到标题：2. CUDA 中的向量化访存类型\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2. CUDA 中的向量化访存类型\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw08.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=KB9OPz3t64E65pRDUwPhKHoWh38%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"3-向量化实现与性能对比\">\u003Ca class=\"anchor\" href=\"#3-向量化实现与性能对比\" aria-label=\"跳转到标题：3. 向量化实现与性能对比\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>3. 向量化实现与性能对比\u003C\u002Fh4>\u003Cul>\n\u003Cli>不同向量化程度的性能测试\u003C\u002Fli>\n\u003Cli>寄存器使用与溢出问题\u003C\u002Fli>\n\u003Cli>数据规模对向量化效果的影响\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3 id=\"半精度计算\">\u003Ca class=\"anchor\" href=\"#半精度计算\" aria-label=\"跳转到标题：半精度计算\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>半精度计算\u003C\u002Fh3>\u003Ch4 id=\"1-半精度浮点数简介\">\u003Ca class=\"anchor\" href=\"#1-半精度浮点数简介\" aria-label=\"跳转到标题：1. 半精度浮点数简介\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>1. 半精度浮点数简介\u003C\u002Fh4>\u003Cul>\n\u003Cli>结构对比（FP16 vs FP32）\u003C\u002Fli>\n\u003Cli>CUDA 中的 half 类型与转换函数\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw09.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=OPnjbsp2zMEcW74nq0LzYJfMRKQ%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch4 id=\"2-半精度性能优势\">\u003Ca class=\"anchor\" href=\"#2-半精度性能优势\" aria-label=\"跳转到标题：2. 半精度性能优势\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>2. 半精度性能优势\u003C\u002Fh4>\u003Cul>\n\u003Cli>内存占用减半\u003C\u002Fli>\n\u003Cli>计算强度提升\u003C\u002Fli>\n\u003Cli>向量化支持（half2）\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch4 id=\"3-向量化计算与访存结合\">\u003Ca class=\"anchor\" href=\"#3-向量化计算与访存结合\" aria-label=\"跳转到标题：3. 向量化计算与访存结合\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>3. 向量化计算与访存结合\u003C\u002Fh4>\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_17_zw10.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=c8YD77sJCsrU3kpMpcTv%2FrG48L4%3D\" alt=\"\">\u003C\u002Fp>\n\u003Ch3 id=\"课后思考题\">\u003Ca class=\"anchor\" href=\"#课后思考题\" aria-label=\"跳转到标题：课后思考题\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>课后思考题\u003C\u002Fh3>\u003Cp>1.可以有效使用 float3 吗?如果可以，怎么做?\u003Cbr>2.之前提到向量化访存有多种方法，今天主要介绍了使用内置类型的方式，想想还有什么其他方式吗?(Hint: 今天的课中其实已经提到了，同时也和第一问有关)\u003C\u002Fp>\n\u003Ch3 id=\"总结与思考\">\u003Ca class=\"anchor\" href=\"#总结与思考\" aria-label=\"跳转到标题：总结与思考\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>总结与思考\u003C\u002Fh3>\u003Cp>本节课我们对优化路径进行了系统性学习，向量化访存 → 半精度 → 向量化计算，性能提升幅度对比等知识。后续课程我们将进一步对于优化方法，算子融合、内存层级优化等知识进行学习，同时将会迎来新的挑战！\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_01_zw6.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=Ffj54EQtw2SzZEJdvnMd%2BBMhib0%3D\" alt=\"2025冬季训练营课程表\">\u003C\u002Fp>\n\u003Ch3 id=\"学习贴士\">\u003Ca class=\"anchor\" href=\"#学习贴士\" aria-label=\"跳转到标题：💡 学习贴士\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>💡 学习贴士\u003C\u002Fh3>\u003Cp>更多细节及完整课程内容，请观看直播或查看课程回放。\u003C\u002Fp>\n\u003Cp>链接：\u003Ca href=\"https:\u002F\u002Fwww.infinitensor.com\u002Fcamp\u002Fwinter2025\">https:\u002F\u002Fwww.infinitensor.com\u002Fcamp\u002Fwinter2025\u003C\u002Fa>\u003C\u002Fp>\n\u003Ch3 id=\"训练营报名持续进行中\">\u003Ca class=\"anchor\" href=\"#训练营报名持续进行中\" aria-label=\"跳转到标题：📚 训练营报名持续进行中\">\u003Cspan class=\"octicon octicon-link\">\u003C\u002Fspan>\u003C\u002Fa>📚 训练营报名持续进行中\u003C\u002Fh3>\u003Cp>扫码报名与加入群聊，或点击 \u003Cstrong>「阅读原文」\u003C\u002Fstrong> 直达官网报名\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"https:\u002F\u002Foss.infinitensor.com\u002Fprod\u002Fimages\u002FKnowledge_Base\u002FKB_01_zw7.jpg?Expires=1792752821&OSSAccessKeyId=LTAI5t9R6uzvTa69CT92cFvV&Signature=ZU5%2F7VUywJvItSFp2MjcU%2FW%2FeuQ%3D\" alt=\"\">\u003C\u002Fp>\n",[128,130,132,136,139,142,145,148,151,154,157,160,163,166,169,172,174,177,180,183,185,187,190],{"id":129,"text":129,"level":9},"课程回顾与背景引入",{"id":131,"text":131,"level":9},"性能瓶颈分析与内存墙",{"id":133,"text":134,"level":135},"1-传输开销与计算时间对比","1. 传输开销与计算时间对比",4,{"id":137,"text":138,"level":135},"2-内存墙概念引入","2. 内存墙概念引入",{"id":140,"text":141,"level":9},"roofline-模型-量化性能瓶颈","Roofline 模型：量化性能瓶颈",{"id":143,"text":144,"level":135},"1-roofline-模型基本概念","1. Roofline 模型基本概念",{"id":146,"text":147,"level":135},"2-模型分区解释","2. 模型分区解释",{"id":149,"text":150,"level":135},"3-向量加法案例分析","3. 向量加法案例分析",{"id":152,"text":153,"level":9},"性能分析工具-nsight-compute-ncu","性能分析工具：Nsight Compute（NCU）",{"id":155,"text":156,"level":135},"1-ncu-简介与使用","1. NCU 简介与使用",{"id":158,"text":159,"level":135},"2-关键分析界面","2. 关键分析界面",{"id":161,"text":162,"level":9},"向量化技术-提升访存效率","向量化技术：提升访存效率",{"id":164,"text":165,"level":135},"1-什么是向量化","1. 什么是向量化？",{"id":167,"text":168,"level":135},"2-cuda-中的向量化访存类型","2. CUDA 中的向量化访存类型",{"id":170,"text":171,"level":135},"3-向量化实现与性能对比","3. 向量化实现与性能对比",{"id":173,"text":173,"level":9},"半精度计算",{"id":175,"text":176,"level":135},"1-半精度浮点数简介","1. 半精度浮点数简介",{"id":178,"text":179,"level":135},"2-半精度性能优势","2. 半精度性能优势",{"id":181,"text":182,"level":135},"3-向量化计算与访存结合","3. 向量化计算与访存结合",{"id":184,"text":184,"level":9},"课后思考题",{"id":186,"text":186,"level":9},"总结与思考",{"id":188,"text":189,"level":9},"学习贴士","💡 学习贴士",{"id":191,"text":192,"level":9},"训练营报名持续进行中","📚 训练营报名持续进行中"]