Windows6:CUDA、cuDNN、TensorRT版本

版本匹配 CUDA cuDNN TensorRT 常见报错解决 CMake编译YOLOV5报错：No CUDA toolset found Could not locate zlibwapi.dll Could not load library cudnn_cnn_infer64_8.dll.TensorRT was linked against cuDNN 8.4.1 but loaded cuDNN 8.4.0 TensorRT was linked against cuBLAS/cuBLAS LT 11.8.0 but loaded cuBLAS/cuBLAS LT 11.6.5 版本

版本匹配

CUDA

找到CUDA对应版本，点击进去。点击【Release Notes】查看，这里以CUDA Toolkit v11.6.1为例

NVIDIA Documentation

Reference the latest NVIDIA products, libraries and API documentation. Including CUDA and NVIDIA GameWorks product families.

https://docs.nvidia.com/cuda/archive/

cuDNN

下载cuDNN后缀就是对应的TensorRT版本

TensorRT

版本对应cuDNN的版本，这里以TensorRT 8.4.1为例

Release Notes :: NVIDIA Deep Learning TensorRT Documentation

NVIDIA TensorRT is a C++ library that facilitates high performance inference on NVIDIA GPUs. It is designed to work in connection with deep learning frameworks that are commonly used for training. TensorRT focuses specifically on running an already trained network quickly and efficiently on a GPU for the purpose of generating a result; also known as inferencing.

https://docs.nvidia.com/deeplearning/tensorrt/release-notes/index.html

常见报错解决

CMake编译YOLOV5报错：No CUDA toolset found

将C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.4\extras\visual_studio_integration\MSBuildExtensions 下的四个文件，拷贝到以下路径：C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\MSBuild\Microsoft\VC\v160\BuildCustomizations ，再编译，通过。

Could not locate zlibwapi.dll

下载

ZLIB DLL Home Page

ZLIB is a compression library compatible with the gzip format. It has been written by Jean-Loup Gailly and Mark Adler. It is free software. The official page is there, but if you are interested only in Windows 95/98/NT/2000/XP/2003 support, you have found the right place.

http://www.winimage.com/zLibDll/

选择zlib123dll.zip.

解压文件

zlibwapi.lib 放在C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.4\lib

zlibwapi.dll 放在 C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.4\bin

Could not load library cudnn_cnn_infer64_8.dll.

TensorRT was linked against cuDNN 8.4.1 but loaded cuDNN 8.4.0

在安装了zlip之后仍然出现该问题，最有可能是安装的CUDNN版本与电脑支持的CUDNN版本不一致

可在https://developer.nvidia.com/rdp/cudnn-archive重新下载对应的版本。把里面的bin、lib、include文件复制放在cuDNN对应bin、lib、include目录里面。C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.4

TensorRT was linked against cuBLAS/cuBLAS LT 11.8.0 but loaded cuBLAS/cuBLAS LT 11.6.5

这是因为安装的cuda是11.6.5版本的，里面的cublas不是11.8.0，安装11.8.0版本的cuda就好了，我们可以看一下cuda11.3.1的介绍

Release Notes :: CUDA Toolkit Documentation

The Release Notes for the CUDA Toolkit. The release notes for the NVIDIA® CUDA® Toolkit can be found online at https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html. Note: The release notes have been reorganized into two major sections: the general CUDA release notes, and the CUDA libraries release notes including historical information for 11.x releases.

https://docs.nvidia.com/cuda/archive/11.6.1/cuda-toolkit-release-notes/index.html

我当时安装的是CUDA11.4.4，我又查了一下11.4.4对应的cuBLAS，果然是11.6.5👉，Nvidia更新驱动CUDA，没有更新对应的组件😂，重新安装CUDA11.6.1

版本

CUDA	cuDNN	TensorRT	Visual Studio	cuBLAs	Torch
11.7/11.6	8.4.1	8.4.1	2022(11.7)	11.10.1.25	1.12.0
ㅤ	ㅤ	ㅤ	ㅤ	ㅤ	ㅤ