什么是cuDNN?如何安装CUDA和cuDNN-赛事专区-网游活动专题站

什么是cuDNN?如何安装CUDA和cuDNN

赛事专区 · 2026-07-11 02:48:42

文章原文指路：https://juejin.cn/post/7479993915041660968

正文如下：

cuDNN是什么?为什么要安装cuDNN?本文将介绍nvidia硬件和驱动(包含nvidia driver),cuda工具包(cuda toolkit),cuDNN系列库和TensorRT,讲解不同层次硬件和驱动以及软件的关系和作用.并使用腾讯cloud stuio做示例,并安装和配置pytorch的GPU加速.

cloud studio介绍Cloud Studio（云端 IDE）是基于浏览器的集成式开发环境，为开发者提供了一个稳定的云端工作站。支持CPU与GPU的访问。用户在使用 Cloud Studio 时无需安装，随时随地打开浏览器即可使用。 Cloud Studio支持免费的CPU环境(每月5w mins)和免费的GPU环境(一张Tesla T4 16G)(每月1w mins).本文将用Cloud Studio的GPU环境演示说明.

开启Cloud Studio GPU空间首先注册并开启Cloud Studio,点击链接curl.qcloud.com/sdeIX8nx点击ide.cloud.tencent.com/ 到Cloud Studio主页面如下图,点击空间模版 -> AI模版 -> Pytorch2.0.0 截屏2025-03-10 20.01.11.png选择免费基础版 -> 确认截屏2025-03-10 20.02.48.png点击高性能工作空间. Pytorch2.0.0 gssrak这个就是已经创建的GPU空间了.可以看到这里已经有绿色圆点,并显示运行中. 截屏2025-03-10 20.03.57.png点击Pytorch2.0.0 gssrak进入空间,等待不到一分钟则会加载完成截屏2025-03-10 20.06.39.pngNvidia driverNvidia Driver是专为nvidia GPU的驱动程序.有了Nvidia Drvier,才可以正确驱动GPU,从而正常输出显示画面(针对studio专业显卡或者游戏显卡)和加速科学计算(针对数据中心显卡等).它也是之后安装CUDA toolkit或者cuDNN的基础.

由于Cloud Studio基于容器技术,已经在宿主机和GPU工作空间(本质是容器)安装了同一版本的Nvidia Driver.我们可以使用nvidia-smi查看

打开终端,输入nvidia-smi:代码语言：txt复制(base) root@VM-24-95-ubuntu:/workspace# nvidia-smi

Mon Mar 10 12:13:25 2025

+-----------------------------------------------------------------------------+

| NVIDIA-SMI 525.105.17 Driver Version: 525.105.17 CUDA Version: 12.0 |

|-------------------------------+----------------------+----------------------+

| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |

| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |

| | | MIG M. |

|===============================+======================+======================|

| 0 Tesla T4 On | 00000000:00:09.0 Off | 0 |

| N/A 31C P8 10W / 70W | 2MiB / 15360MiB | 0% Default |

| | | N/A |

+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+

| Processes: |

| GPU GI CI PID Type Process name GPU Memory |

| ID ID Usage |

|=============================================================================|

| No running processes found |

+-----------------------------------------------------------------------------+

Driver Version: 525.105.17 指Nvidia Driver版本是525.105.17

CUDA Version: 12.0 指目前的Nvidia Driver版本所能支持的最高 CUDA版本是12.0

也就是此时机器支持CUDA12.0以及 <= CUDA12.0的其他版本(CUDA11.8, CUDA11.7, CUDA10.0 等).另一方面 CUDA12.1, CUDA12.8等高于 CUDA12.0的版本,则不被支持.

CUDA toolkitCUDA Toolkit 是 NVIDIA 提供的一套完整的开发工具集，用于开发和优化 CUDA 程序.它包括编译器（如 nvcc）、调试器、运行时库(cudart)、性能分析工具以及各种数学和计算库. 注意如果只需要运行tensotflow或pytorch其实不需要安装(完全版) CUDA toolkit,在安装pytorch或者tensorflow时候自带的cuDNN的子集既可实现GPU加速计算.近在需要开发CUDA算子,编译GPU加速实现(如Apex库)等情况下需要安装CUDA toolkit

Cloud Studio已经默认安装配置了CUDA toolkit 版本11.7

nvcc-V查看是否安装了CUDA toolkit代码语言：txt复制(base) root@VM-24-95-ubuntu:/workspace# nvcc -V

nvcc: NVIDIA (R) Cuda compiler driver

Built on Wed_Jun__8_16:49:14_PDT_2022

Cuda compilation tools, release 11.7, V11.7.99

Build cuda_11.7.r11.7/compiler.31442593_0 echo $PATH,检查是否包含过了路径/usr/local/cuda/bin

代码语言：txt复制(base) root@VM-24-95-ubuntu:/workspace# echo $PATH

/etc/.hai/cloud_studio/vendor/modules/code-oss-dev/bin/remote-cli:/root/miniforge3/bin:/root/miniforge3/condabin:/root/miniforge3/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin

echo $LD_LIBRARY_PATH,检查是否包含过了路径/usr/local/cuda/lib64

代码语言：txt复制 (base) root@VM-24-95-ubuntu:/workspace# echo $LD_LIBRARY_PATH

/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64

cuDNNcuDNN介绍NVIDIA CUDA 深度神经网络库（cuDNN) 是一个 GPU 加速的深度神经网络基本操作库。它提供了深度神经网络（DNN）应用中频繁出现的运算的优化实现.cuDNN是实际在tensorflow,pytorch或大模型部署平台的GPU加速的实现.

ref: 官方网站:docs.nvidia.com/cudnn/index…官方文档:docs.nvidia.com/deeplearnin…官方安装linux下的cuDNN:docs.nvidia.com/deeplearnin…此时如果按照如上所述使用Pytorch2.0.0空间模版则不需要另外再安装cuDNN.因为此时Cloud Studio已经安装并配置好了GPU版本的pytorch,也就是说需要的cuDNN的子集.

查看cuDNN版本查看pytorch是否可以调用cudapython -c "import torch;print(torch.cuda.is_available())"

查看cuDNN是否启用python -c "import torch;print(torch.backends.cudnn.enabled)"

查看cuDNN版本python -c "import torch;print(torch.backends.cudnn.version())"

代码语言：txt复制(base) root@VM-24-95-ubuntu:/workspace# python -c "import torch;print(torch.cuda.is_available())"

True

(base) root@VM-24-95-ubuntu:/workspace# python -c "import torch;print(torch.backends.cudnn.enabled)"

True

(base) root@VM-24-95-ubuntu:/workspace# python -c "import torch;print(torch.backends.cudnn.version())"

8500

因为是pytorch自带的cuDNN的子集,使用代码查看so库find $(python -c "import torch; print(torch.__path__[0])") -name "*cudnn*so*"

代码语言：txt复制(base) root@VM-24-95-ubuntu:/workspace# find $(python -c "import torch; print(torch.__path__[0])") -name "*cudnn*so*"

/root/miniforge3/lib/python3.10/site-packages/torch/lib/libcudnn.so.8

/root/miniforge3/lib/python3.10/site-packages/torch/lib/libcudnn_adv_infer.so.8

/root/miniforge3/lib/python3.10/site-packages/torch/lib/libcudnn_cnn_train.so.8

/root/miniforge3/lib/python3.10/site-packages/torch/lib/libcudnn_adv_train.so.8

/root/miniforge3/lib/python3.10/site-packages/torch/lib/libcudnn_ops_train.so.8

/root/miniforge3/lib/python3.10/site-packages/torch/lib/libcudnn_cnn_infer.so.8

/root/miniforge3/lib/python3.10/site-packages/torch/lib/libcudnn_ops_infer.so.8

验证cuDNN安装安装示例文件和依赖apt -y install libcudnn8-samples libfreeimage-dev build-essential由于刚刚看Cloud Studio的pytorch自带的cuDNN是8500版本所以此处安装libcudnn8-samples.编译cd /usr/src/cudnn_samples_v8/mnistCUDNN && make clean && make运行./mnistCUDNN 出现Test passed!则为安装cuDNN成功. logs of `./mnistCUDNN`

代码语言：javascript复制bash 代码解读复制代码(base) root@VM-24-95-ubuntu:/usr/src/cudnn_samples_v8/mnistCUDNN# make clean && make

rm -rf *o

rm -rf mnistCUDNN

CUDA_VERSION is 11070

Linking agains cublasLt = true

CUDA VERSION: 11070

TARGET ARCH: x86_64

HOST_ARCH: x86_64

TARGET OS: linux

SMS: 35 50 53 60 61 62 70 72 75 80 86 87

/usr/local/cuda/bin/nvcc -I/usr/local/cuda/include -I/usr/local/cuda/include -IFreeImage/include -ccbin g++ -m64 -gencode arch=compute_35,code=sm_35 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_53,code=sm_53 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_62,code=sm_62 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_72,code=sm_72 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_87,code=sm_87 -gencode arch=compute_87,code=compute_87 -o fp16_dev.o -c fp16_dev.cu

nvcc warning : The 'compute_35', 'compute_37', 'sm_35', and 'sm_37' architectures are deprecated, and may be removed in a future release (Use -Wno-deprecated-gpu-targets to suppress warning).

g++ -I/usr/local/cuda/include -I/usr/local/cuda/include -IFreeImage/include -o fp16_emu.o -c fp16_emu.cpp

g++ -I/usr/local/cuda/include -I/usr/local/cuda/include -IFreeImage/include -o mnistCUDNN.o -c mnistCUDNN.cpp

/usr/local/cuda/bin/nvcc -ccbin g++ -m64 -gencode arch=compute_35,code=sm_35 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_53,code=sm_53 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_62,code=sm_62 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_72,code=sm_72 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_87,code=sm_87 -gencode arch=compute_87,code=compute_87 -o mnistCUDNN fp16_dev.o fp16_emu.o mnistCUDNN.o -I/usr/local/cuda/include -I/usr/local/cuda/include -IFreeImage/include -L/usr/local/cuda/lib64 -L/usr/local/cuda/lib64 -L/usr/local/cuda/lib64 -lcublasLt -LFreeImage/lib/linux/x86_64 -LFreeImage/lib/linux -lcudart -lcublas -lcudnn -lfreeimage -lstdc++ -lm

nvcc warning : The 'compute_35', 'compute_37', 'sm_35', and 'sm_37' architectures are deprecated, and may be removed in a future release (Use -Wno-deprecated-gpu-targets to suppress warning).

(base) root@VM-24-95-ubuntu:/usr/src/cudnn_samples_v8/mnistCUDNN# ./mnistCUDNN

Executing: mnistCUDNN

cudnnGetVersion() : 8500 , CUDNN_VERSION from cudnn.h : 8500 (8.5.0)

Host compiler version : GCC 9.4.0

There are 1 CUDA capable devices on your machine :

device 0 : sms 40 Capabilities 7.5, SmClock 1590.0 Mhz, MemSize (Mb) 14928, MemClock 5001.0 Mhz, Ecc=1, boardGroupID=0

Using device 0

Testing single precision

Loading binary file data/conv1.bin

Loading binary file data/conv1.bias.bin

Loading binary file data/conv2.bin

Loading binary file data/conv2.bias.bin

Loading binary file data/ip1.bin

Loading binary file data/ip1.bias.bin

Loading binary file data/ip2.bin

Loading binary file data/ip2.bias.bin

Loading image data/one_28x28.pgm

Performing forward propagation ...

Testing cudnnGetConvolutionForwardAlgorithm_v7 ...

^^^^ CUDNN_STATUS_SUCCESS for Algo 1: -1.000000 time requiring 0 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 0: -1.000000 time requiring 0 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 2: -1.000000 time requiring 57600 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 5: -1.000000 time requiring 178432 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 7: -1.000000 time requiring 2057744 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 4: -1.000000 time requiring 184784 memory

^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory

^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory

Testing cudnnFindConvolutionForwardAlgorithm ...

^^^^ CUDNN_STATUS_SUCCESS for Algo 0: 0.027136 time requiring 0 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 1: 0.027680 time requiring 0 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 2: 0.059392 time requiring 57600 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 5: 0.095232 time requiring 178432 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 7: 0.149504 time requiring 2057744 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 4: 5.357568 time requiring 184784 memory

^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory

^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory

Testing cudnnGetConvolutionForwardAlgorithm_v7 ...

^^^^ CUDNN_STATUS_SUCCESS for Algo 4: -1.000000 time requiring 2450080 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 7: -1.000000 time requiring 1433120 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 1: -1.000000 time requiring 0 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 0: -1.000000 time requiring 0 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 2: -1.000000 time requiring 128000 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 5: -1.000000 time requiring 4656640 memory

^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory

^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory

Testing cudnnFindConvolutionForwardAlgorithm ...

^^^^ CUDNN_STATUS_SUCCESS for Algo 1: 0.088064 time requiring 0 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 0: 0.088352 time requiring 0 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 5: 0.129024 time requiring 4656640 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 7: 0.135936 time requiring 1433120 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 2: 0.144864 time requiring 128000 memory

^^^^ CUDNN_STATUS_SUCCESS for Algo 4: 5.752384 time requiring 2450080 memory

^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 6: -1.000000 time requiring 0 memory

^^^^ CUDNN_STATUS_NOT_SUPPORTED for Algo 3: -1.000000 time requiring 0 memory

Resulting weights from Softmax:

0.0000000 0.9999399 0.0000000 0.0000000 0.0000561 0.0000000 0.0000012 0.0000017 0.0000010 0.0000000

Loading image data/three_28x28.pgm