双4090显卡之间带宽 Nvidia官方测试用例

====测试双4090显卡之间带宽<Nvidia官方测试用例>====

参考李沐视频：单卡、多卡 BERT、GPT2 训练性能【100亿模型计划】_哔哩哔哩_bilibili

参考李沐项目：GitHub - mli/transformers-benchmarks: real Transformer TeraFLOPS on various GPUs

参考他人测试：https://gist.github.com/joshlk/bbb1aca6e70b11d251886baee6423dcb

参考具体项目：cuda-samples/Samples/5_Domain_Specific/p2pBandwidthLatencyTest at master · NVIDIA/cuda-samples · GitHub

Nvidia官方总项目地址：GitHub - NVIDIA/cuda-samples: Samples for CUDA Developers which demonstrates features in CUDA Toolkit

Nvidia官方总项目下载地址：https://github.com/NVIDIA/cuda-samples.git

==首先查看双4090显卡之间连接==

$ nvidia-smi topo -m

再实施思路：下载源代码->编译程序->执行

==下载==

$ git clone https://github.com/NVIDIA/cuda-samples.git //下载总项目

$ sudo apt install freeglut3-dev build-essential libx11-dev libxmu-dev libxi-dev libgl1-mesa-glx libglu1-mesa libglu1-mesa-dev libglfw3-dev libgles2-mesa-dev //安装可能用到的依赖包

==编译==

$ cd ~/cuda-samples/Samples/5_Domain_Specific/p2pBandwidthLatencyTest //进入到测试项目文件夹

$ make //编译程序

==执行==

$ cd ~/cuda-samples/Samples/5_Domain_Specific/p2pBandwidthLatencyTest; ./p2pBandwidthLatencyTest //执行

or

$ cd ~/cuda-samples/bin/x86_64/linux/release; ./p2pBandwidthLatencyTest //执行