当前位置： 首页 > news >正文

【DNN】基础环境搭建 - 指南

news 2026/3/26 19:23:43

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

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
    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
    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）
    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
    在这里插入图片描述
  • 显卡信息
    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

# 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

    // 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:
      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:
      ./test_cuda

11.8.0

在这里插入图片描述

# 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

在这里插入图片描述

# 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

在这里插入图片描述

# 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

# 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 ${CUDNN_INC_DIR}/cudnn_version.h | grep CUDNN_MAJOR -A 2
  • 编译运行测试程序
    // test_cudnn.cpp
#include <cudnn.h>#include <iostream>#include <cstdlib>int main() {cudnnHandle_t handle;cudnnStatus_t status = cudnnCreate(&handle);if (status != CUDNN_STATUS_SUCCESS) {std::cerr << "cuDNN初始化失败: " << cudnnGetErrorString(status) << std::endl;return EXIT_FAILURE;}std::cout << "✓ cuDNN初始化成功" << std::endl;// 获取版本信息size_t version = cudnnGetVersion();std::cout << "✓ cuDNN版本: " << version << std::endl;cudnnDestroy(handle);std::cout << "✓ cuDNN测试完成: 所有操作成功" << std::endl;return EXIT_SUCCESS;}
    • 编译
      g++ -o test_cudnn test_cudnn.cpp -I${CUDA_INC_DIR} -I${CUDA_CUPTI_INC_DIR} -I${CUDNN_INC_DIR} -L${CUDNN_LIB_DIR} -lcudnn
    • 运行
      ./test_cudnn

8.9.7.29_cuda11

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

# 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

# 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

# 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)

在这里插入图片描述

# 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)

在这里插入图片描述

# 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)

在这里插入图片描述

# 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
    # 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
    # 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
http://www.jsqmd.com/news/35689/

相关文章：

  • 解码lvgl图片
  • 句
  • 从功能测试到自动化测试开发：软件测试工程师技能提升指南
  • 2025年质量好的进口品牌平面铰链行业内知名厂家排行榜
  • 2025年质量好的冷凝式衣物烘干机TOP实力厂家推荐榜
  • 2025年比较好的包边净化铝型材高评价厂家推荐榜
  • JSON 学习笔记
  • 2025年热门的衣柜抽屉滑轨厂家实力及用户口碑排行榜
  • 2025年口碑好的直角中空旋转平台TOP实力厂家推荐榜
  • 深入解析：场景美术师的“无限画板”：UE5中非破坏性的材质混合(Material Blending)工作流
  • 2025年热门的广播音响厂家最新权威推荐排行榜
  • 使用AdGuard屏蔽52破解置顶帖
  • 2025年知名的反弹缓冲滑轨行业内知名厂家排行榜
  • 网络攻防实战 lab05 靶机 VulnHub IndiShell Lab: Billu_b0x
  • 20251109 之所思 - 人生如梦
  • GitHub 快速入门指南，新手必备的高效使用手册！
  • 2025年靠谱的工地铺路钢板租赁行业内口碑厂家排行榜
  • MySQL索引（四）：深入剖析索引失效的原因与优化方案
  • 实用指南：Node.js模块化开发实训案例
  • PHPinclude-labs-level 0-11 WP
  • AI元人文：从“真理宫殿”到“可能性土壤”的哲学升华
  • 002 vue3-admin项目的目录及文件说明之.npmrc文件
  • 2025年口碑好的T恤定制厂家推荐及采购参考
  • Aspire开启云原生开发新纪元：微软推出多语言应用开发平台
  • Matlab实现基于Matrix Pencil算法实现声源信号角度和时间估计
  • 2025年口碑好的不锈钢门吸厂家实力及用户口碑排行榜
  • 2025年靠谱的三节同步5D滑轨厂家最新推荐排行榜
  • Oracle数据库空间深度回收：从诊断到优化实战指南
  • 2025年比较好的变频器安装用户好评厂家排行
  • 2025年质量好的TC4钛棒厂家最新用户好评榜