Cuda 驱动安装

Author: ACatSmiling

Since: 2024-11-13

CUDA（Compute Unified Device Architecture）：是 NVIDIA 推出的一种并行计算平台和编程模型，它允许开发者利用 NVIDIA GPU（图形处理器）的强大计算能力进行通用计算，而不仅仅局限于图形处理。简单来说，CUDA 提供了一种方式，让程序员可以像编写 CPU 程序一样来编写能够在 GPU 上高效运行的程序。

安装步骤：

  1. 根据操作系统版本和 GPU 型号等，下载对应的 cuda 驱动，下载地址：https://developer.nvidia.com/cuda-downloads

    image-20241114230828037

  2. 检查服务器 GPU 识别情况，确保 GPU 卡能够完全识别，如不能识别，需要进行重新插拔、对调测试等步骤进行硬件排查。

    [root@desong-gpu-wushan ~]# lspci | grep -i nvidia
00:07.0 3D controller: NVIDIA Corporation Device 2236 (rev a1)

  3. 安装 gcc，g++ 编译器。（GPU 驱动安装时，需要 gcc 编译器）

    [root@desong-gpu-wushan ~]# gcc -v
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/usr/libexec/gcc/x86_64-redhat-linux/4.8.5/lto-wrapper
Target: x86_64-redhat-linux
Configured with: ../configure --prefix=/usr --mandir=/usr/share/man --infodir=/usr/share/info --with-bugurl=http://bugzilla.redhat.com/bugzilla --enable-bootstrap --enable-shared --enable-threads=posix --enable-checking=release --with-system-zlib --enable-__cxa_atexit --disable-libunwind-exceptions --enable-gnu-unique-object --enable-linker-build-id --with-linker-hash-style=gnu --enable-languages=c,c++,objc,obj-c++,java,fortran,ada,go,lto --enable-plugin --enable-initfini-array --disable-libgcj --with-isl=/builddir/build/BUILD/gcc-4.8.5-20150702/obj-x86_64-redhat-linux/isl-install --with-cloog=/builddir/build/BUILD/gcc-4.8.5-20150702/obj-x86_64-redhat-linux/cloog-install --enable-gnu-indirect-function --with-tune=generic --with-arch_32=x86-64 --build=x86_64-redhat-linux
Thread model: posix
gcc version 4.8.5 20150623 (Red Hat 4.8.5-16) (GCC) 
[root@desong-gpu-wushan ~]# g++ -v
Using built-in specs.
COLLECT_GCC=g++
COLLECT_LTO_WRAPPER=/usr/libexec/gcc/x86_64-redhat-linux/4.8.5/lto-wrapper
Target: x86_64-redhat-linux
Configured with: ../configure --prefix=/usr --mandir=/usr/share/man --infodir=/usr/share/info --with-bugurl=http://bugzilla.redhat.com/bugzilla --enable-bootstrap --enable-shared --enable-threads=posix --enable-checking=release --with-system-zlib --enable-__cxa_atexit --disable-libunwind-exceptions --enable-gnu-unique-object --enable-linker-build-id --with-linker-hash-style=gnu --enable-languages=c,c++,objc,obj-c++,java,fortran,ada,go,lto --enable-plugin --enable-initfini-array --disable-libgcj --with-isl=/builddir/build/BUILD/gcc-4.8.5-20150702/obj-x86_64-redhat-linux/isl-install --with-cloog=/builddir/build/BUILD/gcc-4.8.5-20150702/obj-x86_64-redhat-linux/cloog-install --enable-gnu-indirect-function --with-tune=generic --with-arch_32=x86-64 --build=x86_64-redhat-linux
Thread model: posix
gcc version 4.8.5 20150623 (Red Hat 4.8.5-16) (GCC)
    • yum install gcc和yum install gcc-c++。

  4. 安装 kernel-devel 和 kernel-headers 软件包。

    [root@desong-gpu-wushan ~]# yum list | grep kernel
Repodata is over 2 weeks old. Install yum-cron? Or run: yum makecache fast
kernel.x86_64                             3.10.0-693.el7               @anaconda
kernel-abi-whitelists.noarch              3.10.0-693.el7               @base-52 
kernel-devel.x86_64                       3.10.0-693.el7               @anaconda
kernel-headers.x86_64                     3.10.0-693.el7               @anaconda
kernel-tools.x86_64                       3.10.0-693.el7               @anaconda
kernel-tools-libs.x86_64                  3.10.0-693.el7               @anaconda
abrt-addon-kerneloops.x86_64              2.1.11-60.el7.centos         base     
erlang-kernel.x86_64                      R16B-03.18.el7               epel     
kernel.x86_64                             3.10.0-1160.42.2.el7         updates  
kernel-abi-whitelists.noarch              3.10.0-1160.42.2.el7         updates  
kernel-debug.x86_64                       3.10.0-1160.42.2.el7         updates  
kernel-debug-devel.x86_64                 3.10.0-1160.42.2.el7         updates  
kernel-devel.x86_64                       3.10.0-1160.42.2.el7         updates  
kernel-doc.noarch                         3.10.0-1160.42.2.el7         updates  
kernel-headers.x86_64                     3.10.0-1160.42.2.el7         updates  
kernel-tools.x86_64                       3.10.0-1160.42.2.el7         updates  
kernel-tools-libs.x86_64                  3.10.0-1160.42.2.el7         updates  
kernel-tools-libs-devel.x86_64            3.10.0-1160.42.2.el7         updates  
libreport-plugin-kerneloops.x86_64        2.1.11-53.el7.centos         base     
lirc-disable-kernel-rc.noarch             0.10.0-16.el7                epel     
php-symfony-http-kernel.noarch            2.8.12-2.el7                 epel     
texlive-l3kernel.noarch                   2:svn29409.SVN_4469-45.el7   base     
texlive-l3kernel-doc.noarch               2:svn29409.SVN_4469-45.el7   base
    • yum install kernet-devel-$(uname -r) kernel-headers-$(uname -r)。

  5. 禁用系统自带的 nouveau 模块。

    # 检查 nouveau 模块是否加载
[root@desong-gpu-wushan ~]# lsmod | grep nouveau
nouveau              1622010  0 
mxm_wmi                13021  1 nouveau
wmi                    19070  2 mxm_wmi,nouveau
video                  24520  1 nouveau
i2c_algo_bit           13413  1 nouveau
drm_kms_helper        159169  2 cirrus,nouveau
ttm                    99345  2 cirrus,nouveau
drm                   370825  5 ttm,drm_kms_helper,cirrus,nouveau
i2c_core               40756  5 drm,i2c_piix4,drm_kms_helper,i2c_algo_bit,nouveau

# 编辑 /usr/lib/modprobe.d/dist-blacklist.conf，添加以下内容，如果没有则创建
[root@desong-gpu-wushan modprobe.d]# vim /usr/lib/modprobe.d/dist-blacklist.conf
blacklist nouveau
options nouveau modeset=0

# 使内核生效
[root@desong-gpu-wushan modprobe.d]# dracut -force

  6. 修改系统运行级别为文本模式，GPU 驱动安装必须在文本模式下进行。

    [root@desong-gpu-wushan modprobe.d]# systemctl set-default multi-user.target
Removed symlink /etc/systemd/system/default.target.
Created symlink from /etc/systemd/system/default.target to /usr/lib/systemd/system/multi-user.target.
# 重启系统
[root@desong-gpu-wushan modprobe.d]# reboot

  7. 验证 nouveau 模块禁用生效，并进入文本模式。

    [root@desong-gpu-wushan ~]# lsmod | grep nouveau

  8. cuda 安装。

    # 赋予 cuda 安装包可执行权限
[root@desong-gpu-wushan ~]# chmod +x cuda_12.2.0_535.54.03_linux.run
# 执行 cuda 安装命令
[root@desong-gpu-wushan ~]# ./cuda_12.2.0_535.54.03_linux.run --no-opengl-libs

    image-20241113164756822

    image-20241113164802852

    • 默认全部安装，如果已经安装过 GPU 驱动，则取消 Driver 选项的勾选。
    Driver:   Installed
Toolkit:  Installed in /usr/local/cuda-12.2/

Please make sure that

 -   PATH includes /usr/local/cuda-12.2/bin
 -   LD_LIBRARY_PATH includes /usr/local/cuda-12.2/lib64, or, add /usr/local/cuda-12.2/lib64 to /etc/ld.so.conf and run ldconfig as root

To uninstall the CUDA Toolkit, run cuda-uninstaller in /usr/local/cuda-12.2/bin
To uninstall the NVIDIA Driver, run nvidia-uninstall
Logfile is /var/log/cuda-installer.log

  9. 配置环境变量。

    [root@desong-gpu-wushan ~]# vim /etc/profile
export PATH=/usr/local/cuda/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH

  10. 生效环境变量。

    [root@desong-gpu-wushan ~]# source /etc/profile

  11. 测试 cuda 是否安装成功。

    [root@desong-gpu-wushan ~]# nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Built on Tue_Jun_13_19:16:58_PDT_2023
Cuda compilation tools, release 12.2, V12.2.91
Build cuda_12.2.r12.2/compiler.32965470_0

  12. 修改系统运行级别为图形模式，然后重启系统。

    [root@desong-gpu-wushan ~]# systemctl set-default graphical.target
Removed symlink /etc/systemd/system/default.target.
Created symlink from /etc/systemd/system/default.target to /usr/lib/systemd/system/graphical.target.
[root@desong-gpu-wushan ~]# reboot

  13. 验证GPU驱动及CUDA安装成功，以及图形化界面显示正常。

    [root@desong-gpu-wushan ~]# nvidia-smi
Wed Nov  6 07:18:06 2024       
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.54.03              Driver Version: 535.54.03    CUDA Version: 12.2     |
|-----------------------------------------+----------------------+----------------------+
| 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 A10                     Off | 00000000:00:07.0 Off |                    0 |
|  0%   39C    P0              58W / 150W |      4MiB / 23028MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
                                                                                         
+---------------------------------------------------------------------------------------+
| Processes:                                                                            |
|  GPU   GI   CI        PID   Type   Process name                            GPU Memory |
|        ID   ID                                                             Usage      |
|=======================================================================================|
|  No running processes found                                                           |
+---------------------------------------------------------------------------------------+
posted @ 2024-11-14 23:14  ACatSmiling  阅读(6)  评论(0)  编辑  收藏  举报