【DNN】基础环境搭建

前言

实现CUDA，CUDNN与TensorRT各个版本之间的依赖关系尤为重要，但是在不同的工作环境下可能需要使用不同的版本匹配。本文主要通过软连接的方式实现各个版本之间的自由搭配。

系统信息

操作系统

lsb_release

bash 复制代码

LSB Version:	core-11.1.0ubuntu4-noarch:security-11.1.0ubuntu4-noarch
Distributor ID:	Ubuntu
Description:	Ubuntu 22.04.5 LTS
Release:	22.04
Codename:	jammy

hostnamectl

bash 复制代码

 Static hostname: msi
       Icon name: computer-desktop
         Chassis: desktop
      Machine ID: 0905fc3742014849a8f6e66a18eec86a
         Boot ID: 946b89d00b014454b09cbb0f267b287a
Operating System: Ubuntu 22.04.5 LTS                
          Kernel: Linux 6.8.0-85-generic
    Architecture: x86-64
 Hardware Vendor: Micro-Star International Co., Ltd.
  Hardware Model: MS-7D25

显卡驱动

查看系统推荐的显卡驱动（选择带recommended的驱动）

bash 复制代码

ubuntu-drivers devices

== /sys/devices/pci0000:00/0000:00:01.0/0000:01:00.0 ==
modalias : pci:v000010DEd00001E07sv000019DAsd00001503bc03sc00i00
vendor   : NVIDIA Corporation
model    : TU102 [GeForce RTX 2080 Ti Rev. A]
driver   : nvidia-driver-470-server - distro non-free
driver   : nvidia-driver-550 - distro non-free
driver   : nvidia-driver-545-open - distro non-free
driver   : nvidia-driver-545 - distro non-free
driver   : nvidia-driver-570-open - distro non-free
driver   : nvidia-driver-550-open - distro non-free
driver   : nvidia-driver-535-server - distro non-free
driver   : nvidia-driver-570-server-open - distro non-free
driver   : nvidia-driver-580-server-open - distro non-free
driver   : nvidia-driver-570-server - distro non-free
driver   : nvidia-driver-535-open - distro non-free
driver   : nvidia-driver-535 - distro non-free
driver   : nvidia-driver-580 - distro non-free recommended
driver   : nvidia-driver-470 - distro non-free
driver   : nvidia-driver-450-server - distro non-free
driver   : nvidia-driver-580-server - distro non-free
driver   : nvidia-driver-580-open - distro non-free
driver   : nvidia-driver-570 - distro non-free
driver   : nvidia-driver-535-server-open - distro non-free
driver   : nvidia-driver-418-server - distro non-free
driver   : xserver-xorg-video-nouveau - distro free builtin

安装驱动

显卡信息

bash 复制代码

nvidia-smi

+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 580.65.06              Driver Version: 580.65.06      CUDA Version: 13.0     |
+-----------------------------------------+------------------------+----------------------+
| 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  NVIDIA GeForce RTX 2080 Ti     Off |   00000000:01:00.0  On |                  N/A |
| 44%   51C    P2            127W /  260W |    1118MiB /  11264MiB |     44%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI              PID   Type   Process name                        GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A            1737      G   /usr/lib/xorg/Xorg                      117MiB |
|    0   N/A  N/A            1899      G   /usr/bin/gnome-shell                    104MiB |
|    0   N/A  N/A            3911      C   python                                  872MiB |
+-----------------------------------------------------------------------------------------+

CUDA

https://developer.nvidia.com/cuda-toolkit-archive

配置环境变量

bash 复制代码

# vim ~/.bashrc

export CUDA_HONE=/usr/local/cuda
export CUDA_INC_DIR=${CUDA_HONE}/include
export CUDA_LIB_DIR=${CUDA_HONE}/lib64
export CUDA_BIN_DIR=${CUDA_HONE}/bin
export CUDA_CUPTI_INC_DIR=${CUDA_HONE}/extras/CUPTI/include
export CUDA_CUPTI_LIB_DIR=${CUDA_HONE}/extras/CUPTI/lib64
export PATH=${CUDA_BIN_DIR}${PATH:+:${PATH}}
export LD_LIBRARY_PATH=${CUDA_LIB_DIR}${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
export LD_LIBRARY_PATH=${CUDA_CUPTI_LIB_DIR}${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

验证是否安装成功

nvcc -V
/usr/local/cuda-11.8/extras/demo_suite/bandwidthTest
/usr/local/cuda-11.8/extras/demo_suite/deviceQuery

编译测试程序

cpp 复制代码

// test_cuda.cpp

#include <stdio.h>
#include <stdlib.h>
#include <cuda_runtime.h>

int main() {
    int deviceCount;
    cudaGetDeviceCount(&deviceCount);
    
    printf("找到 %d 个CUDA设备:\n", deviceCount);
    
    for (int i = 0; i < deviceCount; i++) {
        cudaDeviceProp prop;
        cudaGetDeviceProperties(&prop, i);
        printf("设备 %d: %s\n", i, prop.name);
        printf("  Compute Capability: %d.%d\n", prop.major, prop.minor);
        printf("  全局内存: %.2f GB\n", prop.totalGlobalMem / (1024.0 * 1024.0 * 1024.0));
        printf("  CUDA核心: %d\n", prop.multiProcessorCount * prop.maxThreadsPerMultiProcessor);
    }
    
    // 简单的GPU计算测试
    float *d_array, *h_array;
    h_array = (float*)malloc(10 * sizeof(float));
    cudaMalloc(&d_array, 10 * sizeof(float));
    
    cudaMemcpy(d_array, h_array, 10 * sizeof(float), cudaMemcpyHostToDevice);
    cudaMemcpy(h_array, d_array, 10 * sizeof(float), cudaMemcpyDeviceToHost);
    
    cudaFree(d_array);
    free(h_array);
    
    printf("CUDA内存操作测试: 成功!\n");
    return 0;
}

编译:

bash 复制代码

g++ -o test_cuda test_cuda.cpp -I${CUDA_INC_DIR} -I${CUDA_CUPTI_INC_DIR} -L${CUDA_LIB_DIR} -I${CUDA_CUPTI_LIB_DIR} -lcudart

运行:
bash 复制代码
```
./test_cuda
```

11.8.0

bash 复制代码

# 下载链接
wget https://developer.download.nvidia.com/compute/cuda/11.8.0/local_installers/cuda_11.8.0_520.61.05_linux.run

# 安装(取消驱动安装)
sudo sh cuda_11.8.0_520.61.05_linux.run

# 配置软链接
sudo rm -rf /usr/local/cuda
sudo ln -s /usr/local/cuda-11.8 /usr/local/cuda

# 卸载(需重新配置软连接)
sudo rm -rf /usr/local/cuda
sudo ln -s /usr/local/cuda-11.8 /usr/local/cuda
sudo /usr/local/cuda/bin/cuda-uninstaller
sudo rm -rf /usr/local/cuda

12.4.1

bash 复制代码

# 下载链接
wget https://developer.download.nvidia.com/compute/cuda/12.4.1/local_installers/cuda_12.4.1_550.54.15_linux.run

# 安装(取消驱动安装)
sudo sh cuda_12.4.1_550.54.15_linux.run

# 配置软链接
sudo rm -rf /usr/local/cuda
sudo ln -s /usr/local/cuda-12.4 /usr/local/cuda

# 卸载(需重新配置软连接)
sudo rm -rf /usr/local/cuda
sudo ln -s /usr/local/cuda-12.4 /usr/local/cuda
sudo /usr/local/cuda/bin/cuda-uninstaller
sudo rm -rf /usr/local/cuda

12.8.1

bash 复制代码

# 下载链接
wget https://developer.download.nvidia.com/compute/cuda/12.8.1/local_installers/cuda_12.8.1_570.124.06_linux.run

# 安装(取消驱动安装)
sudo sh cuda_12.8.1_570.124.06_linux.run

# 配置软链接
sudo rm -rf /usr/local/cuda
sudo ln -s /usr/local/cuda-12.8 /usr/local/cuda

# 卸载(需重新配置软连接)
sudo rm -rf /usr/local/cuda
sudo ln -s /usr/local/cuda-12.8 /usr/local/cuda
sudo /usr/local/cuda/bin/cuda-uninstaller
sudo rm -rf /usr/local/cuda

CUDNN

https://developer.download.nvidia.cn/compute/cudnn/redist/cudnn/linux-x86_64/

配置环境变量

bash 复制代码

# vim ~/.bashrc

export CUDNN_HOME=/opt/cudnn
export CUDNN_INC_DIR=${CUDNN_HOME}/include
export CUDNN_LIB_DIR=${CUDNN_HOME}/lib
export LD_LIBRARY_PATH=${CUDNN_LIB_DIR}${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

验证安装是否成功

cat ${CUDNN_INC_DIR}/cudnn_version.h | grep CUDNN_MAJOR -A 2

编译运行测试程序

cpp 复制代码

// test_cudnn.cpp

#include <cudnn.h>
#include <iostream>
#include <cstdlib>

int main() {
    cudnnHandle_t handle;
    cudnnStatus_t status = cudnnCreate(&handle);
    
    if (status != CUDNN_STATUS_SUCCESS) {
        std::cerr << "cuDNN初始化失败: " << cudnnGetErrorString(status) << std::endl;
        return EXIT_FAILURE;
    }
    
    std::cout << "✓ cuDNN初始化成功" << std::endl;
    
    // 获取版本信息
    size_t version = cudnnGetVersion();
    std::cout << "✓ cuDNN版本: " << version << std::endl;
    
    cudnnDestroy(handle);
    std::cout << "✓ cuDNN测试完成: 所有操作成功" << std::endl;
    
    return EXIT_SUCCESS;
}

编译

bash 复制代码

g++ -o test_cudnn test_cudnn.cpp -I${CUDA_INC_DIR} -I${CUDA_CUPTI_INC_DIR} -I${CUDNN_INC_DIR} -L${CUDNN_LIB_DIR} -lcudnn

运行
bash 复制代码
```
./test_cudnn
```

8.9.7.29_cuda11

bash 复制代码

 下载
wget https://developer.download.nvidia.cn/compute/cudnn/redist/cudnn/linux-x86_64/cudnn-linux-x86_64-8.9.7.29_cuda11-archive.tar.xz

# 解压
sudo tar -xf cudnn-linux-x86_64-8.9.7.29_cuda11-archive.tar.xz -C /opt

# 配置软链接
sudo rm -rf /opt/cudnn
sudo ln -s /opt/cudnn-linux-x86_64-8.9.7.29_cuda11-archive /opt/cudnn

# 设置文件权限
sudo chmod a+r /opt/cudnn-linux-x86_64-8.9.7.29_cuda11-archive/include/cudnn*
sudo chmod a+r /opt/cudnn-linux-x86_64-8.9.7.29_cuda11-archive/lib/libcudnn*

# 卸载
sudo rm -rf /opt/cudnn-linux-x86_64-8.9.7.29_cuda11-archive
sudo rm -rf /opt/cudnn

9.2.1.18_cuda12

bash 复制代码

# 下载
wget https://developer.download.nvidia.cn/compute/cudnn/redist/cudnn/linux-x86_64/cudnn-linux-x86_64-9.2.1.18_cuda12-archive.tar.xz

# 解压
sudo tar -xf cudnn-linux-x86_64-9.2.1.18_cuda12-archive.tar.xz -C /opt

# 配置软链接
sudo rm -rf /opt/cudnn
sudo ln -s /opt/cudnn-linux-x86_64-9.2.1.18_cuda12-archive /opt/cudnn

# 设置文件权限
sudo chmod a+r /opt/cudnn-linux-x86_64-9.2.1.18_cuda12-archive/include/cudnn*
sudo chmod a+r /opt/cudnn-linux-x86_64-9.2.1.18_cuda12-archive/lib/libcudnn*

# 卸载
sudo rm -rf /opt/cudnn-linux-x86_64-9.2.1.18_cuda12-archive
sudo rm -rf /opt/cudnn

9.14.0.64_cuda12

bash 复制代码

# 下载
wget https://developer.download.nvidia.cn/compute/cudnn/redist/cudnn/linux-x86_64/cudnn-linux-x86_64-9.14.0.64_cuda12-archive.tar.xz

# 解压
sudo tar -xf cudnn-linux-x86_64-9.14.0.64_cuda12-archive.tar.xz -C /opt

# 配置软链接
sudo rm -rf /opt/cudnn
sudo ln -s /opt/cudnn-linux-x86_64-9.14.0.64_cuda12-archive /opt/cudnn

# 设置文件权限
sudo chmod a+r /opt/cudnn-linux-x86_64-9.14.0.64_cuda12-archive/include/cudnn*
sudo chmod a+r /opt/cudnn-linux-x86_64-9.14.0.64_cuda12-archive/lib/libcudnn*

# 卸载
sudo rm -rf /opt/cudnn-linux-x86_64-8.9.7.29_cuda11-archive
sudo rm -rf /opt/cudnn

TensorRT

https://developer.nvidia.com/tensorrt/download

配置环境变量

bash 复制代码

# vim ~/.bashrc

export TENSORRT_HOME=/opt/tensorrt
export TENSORRT_INC_DIR=${TENSORRT_HOME}/include
export TENSORRT_LIB_DIR=${TENSORRT_HOME}/lib
export TENSORRT_BIN_DIR=${TENSORRT_HOME}/bin
export PATH=${TENSORRT_BIN_DIR}${PATH:+:${PATH}}
export LD_LIBRARY_PATH=${TENSORRT_LIB_DIR}${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

验证是否安装成功

ls -l ${TENSORRT_LIB_DIR}/libnvinfer.so*
trtexec --help
trtexec --onnx=${TENSORRT_HOME}/data/mnist/mnist.onnx | grep 'TensorRT version'

8.6.1.6(CUDA 11.x)

bash 复制代码

# 下载
wget https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/secure/8.6.1/tars/TensorRT-8.6.1.6.Linux.x86_64-gnu.cuda-11.8.tar.gz

# 解压
sudo tar -xf TensorRT-8.6.1.6.Linux.x86_64-gnu.cuda-11.8.tar.gz -C /opt

# 改名(便于区分适配不同的CUDA版本)
sudo mv /opt/TensorRT-8.6.1.6 /opt/TensorRT-8.6.1.6-cuda-11.x

# 软链接
sudo rm -rf /opt/tensorrt
sudo ln -s /opt/TensorRT-8.6.1.6-cuda-11.x /opt/tensorrt

# 删除
sudo rm -rf /opt/TensorRT-8.6.1.6-cuda-11.x
sudo rm -rf /opt/tensorrt

10.7.0.23(CUDA 12.0-12.6)

bash 复制代码

# 下载
wget https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.7.0/tars/TensorRT-10.7.0.23.Linux.x86_64-gnu.cuda-12.6.tar.gz

# 解压
sudo tar -xf TensorRT-10.7.0.23.Linux.x86_64-gnu.cuda-12.6.tar.gz -C /opt

# 改名(便于区分适配不同的CUDA版本)
sudo mv /opt/TensorRT-10.7.0.23 /opt/TensorRT-10.7.0.23-cuda-12.0-12.6

# 软链接
sudo rm -rf /opt/tensorrt
sudo ln -s /opt/TensorRT-10.7.0.23-cuda-12.0-12.6 /opt/tensorrt

# 删除
sudo rm -rf /opt/TensorRT-10.7.0.23-cuda-12.0-12.6
sudo rm -rf /opt/tensorrt

10.8.0.43(CUDA 12.x)

bash 复制代码

# 下载
wget https://developer.nvidia.com/downloads/compute/machine-learning/tensorrt/10.8.0/tars/TensorRT-10.8.0.43.Linux.x86_64-gnu.cuda-12.8.tar.gz

# 解压
sudo tar -xf TensorRT-10.8.0.43.Linux.x86_64-gnu.cuda-12.8.tar.gz -C /opt

# 改名(便于区分适配不同的CUDA版本)
sudo mv /opt/TensorRT-10.8.0.43 /opt/TensorRT-10.8.0.43-cuda-12.x

# 软链接
sudo rm -rf /opt/tensorrt
sudo ln -s /opt/TensorRT-10.8.0.43-cuda-12.x /opt/tensorrt

# 删除
sudo rm -rf /opt/TensorRT-10.8.0.43-cuda-12.x
sudo rm -rf /opt/tensorrt

后记

可根据不同的需求调整版本适配，例如：

基于CUDA(11.8)训练与部署CNN网络

bash 复制代码

# CUDA
sudo rm -rf /usr/local/cuda
sudo ln -s /usr/local/cuda-11.8 /usr/local/cuda

# CUDNN
sudo rm -rf /opt/cudnn
sudo ln -s /opt/cudnn-linux-x86_64-8.9.7.29_cuda11-archive /opt/cud

# TensorRT
sudo rm -rf /opt/tensorrt
sudo ln -s /opt/TensorRT-8.6.1.6-cuda-11.x /opt/tensorrt

基于CUDA(12.8)训练与部署千问大模型

bash 复制代码

# CUDA
sudo rm -rf /usr/local/cuda
sudo ln -s /usr/local/cuda-12.8 /usr/local/cuda

# CUDNN
sudo rm -rf /opt/cudnn
sudo ln -s /opt/cudnn-linux-x86_64-9.14.0.64_cuda12-archive /opt/cudnn

# TensorRT
sudo rm -rf /opt/tensorrt
sudo ln -s /opt/TensorRT-10.8.0.43-cuda-12.x /opt/tensorrt

【DNN】基础环境搭建

目录

前言

系统信息

操作系统

显卡驱动

CUDA

配置环境变量

验证是否安装成功

11.8.0

12.4.1

12.8.1

CUDNN

配置环境变量

验证安装是否成功

8.9.7.29_cuda11

9.2.1.18_cuda12

9.14.0.64_cuda12

TensorRT

配置环境变量

验证是否安装成功

8.6.1.6(CUDA 11.x)

10.7.0.23(CUDA 12.0-12.6)

10.8.0.43(CUDA 12.x)

后记