<![CDATA[我的笔记本]]>

<![CDATA[我的笔记本]]>https://blog.csdn.net/bandaoyuzh-cnhttps://blog.csdn.net/<![CDATA[【CUDA】什么是DSL]]>https://blog.csdn.net/bandaoyu/article/details/163863371https://blog.csdn.net/bandaoyu/article/details/163863371bandaoyuTue, 18 Aug 2026 20:49:27 +0800<![CDATA[【CUDA】TileFusion]]>https://blog.csdn.net/bandaoyu/article/details/163804942https://blog.csdn.net/bandaoyu/article/details/163804942bandaoyuMon, 17 Aug 2026 00:55:10 +0800<![CDATA[【翻译】在多GPU上利用动态In-Switch计算加速MoE]]>https://blog.csdn.net/bandaoyu/article/details/162674901https://blog.csdn.net/bandaoyu/article/details/162674901bandaoyuTue, 07 Jul 2026 22:58:32 +0800<![CDATA[大模型训练并行技术理解-DP/TP/PP/SP/EP]]>https://blog.csdn.net/bandaoyu/article/details/162608515https://blog.csdn.net/bandaoyu/article/details/162608515bandaoyuSun, 05 Jul 2026 18:16:32 +0800<![CDATA[【CUDA】cubin是什么]]>https://blog.csdn.net/bandaoyu/article/details/161912367https://blog.csdn.net/bandaoyu/article/details/161912367bandaoyuFri, 12 Jun 2026 05:07:26 +0800>>：直接启动cubin：先加载模块，再它们是同一类 kernel 的两种装载/启动路径。]]><![CDATA[【CUDA】MNNVL和NVLS的关系]]>https://blog.csdn.net/bandaoyu/article/details/161807927https://blog.csdn.net/bandaoyu/article/details/161807927bandaoyuMon, 08 Jun 2026 21:09:11 +0800跨节点NVLink全连接] --- Node2[GPU节点2]endsubgraph B [NVLS加速层]]]><![CDATA[【AMD】HDP（Host Data Path）是什么]]>https://blog.csdn.net/bandaoyu/article/details/161491296https://blog.csdn.net/bandaoyu/article/details/161491296bandaoyuThu, 28 May 2026 18:07:38 +0800 GPU 显存的直接访问通路，让 HCA/CPU 能通过PCI BAR地址空间，直接读 / 写 GPU 的 VRAM，不用走复杂 DMA 或驱动拷贝。]]><![CDATA[【超节点】HSA跨节点直接访存的原理]]>https://blog.csdn.net/bandaoyu/article/details/159252228https://blog.csdn.net/bandaoyu/article/details/159252228bandaoyuTue, 26 May 2026 08:52:33 +0800<![CDATA[【RDMA】CST=Consistency at Target（目标端一致性操作）]]>https://blog.csdn.net/bandaoyu/article/details/161322766https://blog.csdn.net/bandaoyu/article/details/161322766bandaoyuFri, 22 May 2026 20:22:22 +0800<![CDATA[【GDR】 GPU内存和VMM 的RDMA 内存注册|nvidia-peerme]]>https://blog.csdn.net/bandaoyu/article/details/161262658https://blog.csdn.net/bandaoyu/article/details/161262658bandaoyuWed, 20 May 2026 18:51:06 +0800<![CDATA[【CUDA】store/load普通访存 vs 非临时（Non-Temporal）访存]]>https://blog.csdn.net/bandaoyu/article/details/161120665https://blog.csdn.net/bandaoyu/article/details/161120665bandaoyuFri, 15 May 2026 19:51:35 +0800<![CDATA[【NCCL】NCCL Inspector + Prometheus对NCCL实时监控与快速调试]]>https://blog.csdn.net/bandaoyu/article/details/160896573https://blog.csdn.net/bandaoyu/article/details/160896573bandaoyuFri, 08 May 2026 18:01:19 +0800<![CDATA[【GPUDirect 】GPUDirect RDMA (GDR)、GPUDirect Async (GDA)技术是什么？GPUDirect 家族]]>https://blog.csdn.net/bandaoyu/article/details/160316834https://blog.csdn.net/bandaoyu/article/details/160316834bandaoyuMon, 20 Apr 2026 00:19:17 +0800<![CDATA[【CUDA】NVbandwidth：测量 GPU 互连和内存性能的必备工具]]>https://blog.csdn.net/bandaoyu/article/details/160316610https://blog.csdn.net/bandaoyu/article/details/160316610bandaoyuSun, 19 Apr 2026 23:55:18 +0800<![CDATA[【NVSHMEM】PCIe 距离类型(PIX,PXB,PHB,NOD,SYS)和判断]]>https://blog.csdn.net/bandaoyu/article/details/158935043https://blog.csdn.net/bandaoyu/article/details/158935043bandaoyuWed, 11 Mar 2026 22:09:59 +0800<![CDATA[【nvshmem】nvshmem中的DMA buf是什么？]]>https://blog.csdn.net/bandaoyu/article/details/158509124https://blog.csdn.net/bandaoyu/article/details/158509124bandaoyuSat, 28 Feb 2026 18:46:06 +0800<![CDATA[【工具】docx/word转MD:Pandoc]]>https://blog.csdn.net/bandaoyu/article/details/158468638https://blog.csdn.net/bandaoyu/article/details/158468638bandaoyuFri, 27 Feb 2026 18:12:54 +0800<![CDATA[【NCCL】NCCL通信协议:Simple, LL, LL128]]>https://blog.csdn.net/bandaoyu/article/details/158427203https://blog.csdn.net/bandaoyu/article/details/158427203bandaoyuThu, 26 Feb 2026 18:01:58 +0800<![CDATA[【RCCL】RCCL工具]]>https://blog.csdn.net/bandaoyu/article/details/158392204https://blog.csdn.net/bandaoyu/article/details/158392204bandaoyuWed, 25 Feb 2026 20:34:43 +0800<![CDATA[【RDMA】rdma指令]]>https://blog.csdn.net/bandaoyu/article/details/157478974https://blog.csdn.net/bandaoyu/article/details/157478974bandaoyuWed, 28 Jan 2026 21:48:35 +0800