安装 CUDA9.0 + cuDNN8.0

再唠叨之前的准备:

• 1.修改资源镜像，可用脚本 ubuntu修改资源镜像

• 2.系统保持最新
  sudo apt-get update
sudo apt-get upgrade

我的环境：

 
OS:  ubuntu 16.04.3
 
uname -a
Linux scott-Z170X-UD3 4.10.0-37-generic #41~16.04.1-Ubuntu SMP Fri Oct 6 22:42:59 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux
 
gcc --version
gcc (Ubuntu 5.4.0-6ubuntu1~16.04.5) 5.4.0 20160609
 
 
 

敲黑板！

• tensorflow 有CPU 版，GPU 版。

• 不是所有显卡都支持CUDA,可以在这里看到cuda评分，推荐GPU Compute Capability 最少 >3.0

该网址查看 cuda计算能力评分 GPU Compute Capability

---

CUDA官方网址

你可以选择下载自己的平台安装包

我我用的是deb[local]

CUDA9.0 deb local ubuntu x64 下载地址

Installation Instructions:
sudo dpkg -i cuda-repo-ubuntu1604-9-0-local_9.0.176-1_amd64.deb
sudo apt-key add /var/cuda-repo-<version>/7fa2af80.pub
sudo apt-get update
sudo apt-get install cuda

Other installation options are available in the form of meta-packages. For example, to install all the library packages, replace "cuda" with the "cuda-libraries-9-0" meta package. For more information on all the available meta packages click here.

The CUDA Toolkit contains Open-Source Software. The source code can be found here.
The checksums for the installer and patches can be found in Installer Checksums.
For further information, see the Installation Guide for Linux and the CUDA Quick Start Guide.

敲黑板！

推荐先看pdf [Installation Guide for Linux] , [CUDA Quick Start Guide] ，里面介绍了 不同平台的操作区别。
我只介绍ubuntu的安装

• 此时查看 Cuda compilation tools版本,
  NVIDIA CUDA C/C++ Compiler (NVCC)

nvcc --version
提示找不到命令.我当时很奇怪。 之后，我用ubantu16.04的apt 软件库去安装Cuda compilation tools
sudo apt install nvidia-cuda-toolkit

再查询，发现ubuntu库中的cuda是7.5的，不是我这次安装的版本。
果断删除
sudo apt remove nvidia-cuda-toolkit

• 上网找了资料，发现原来是需要更新环境变量

  • 编辑.bashrc： sudo gedit .bashrc
  • 添加cuda文件路径

  export PATH=/usr/local/cuda-9.0/bin:$PATH
  export LD_LIBRARY_PATH=/usr/local/cuda-9.0/bin/lib64:$LD_LIBRARY_PATH

  • 需要更新环境变量，终端中输入： source ~/.bashrc

原理点此链接 在 Ubuntu 系统中有三种设置环境变量 PATH 的方法

• 再查询，是9.0！到此CUDA9.0 安装成功

---

NVIDIA cuDNN is a GPU-accelerated library of primitives for deep neural networks.

[Download cuDNN v7.0.3 (Sept 28, 2017), for CUDA 9.0] (https://developer.nvidia.com/rdp/cudnn-download)

不同平台操作看guide

windows:
安装完CUDA，直接复制zip包下的文件到 CUDA_PATH

CUDA_PATH  C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v9.0

---

验证驱动版

cat /proc/driver/nvidia/version

查看 Cuda compilation tools版本

nvcc --version或者nvcc -V

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2017 NVIDIA Corporation
Built on Fri_Sep__1_21:08:03_CDT_2017
Cuda compilation tools, release 9.0, V9.0.176

查看GPU 当前信息

nvidia-smi

Fri Oct 20 16:21:13 2017       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 384.81                 Driver Version: 384.81                    |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  GeForce GTX 950     Off  | 00000000:01:00.0  On |                  N/A |
| 33%   33C    P8    12W / 100W |    332MiB /  1995MiB |      6%      Default |
+-------------------------------+----------------------+----------------------+
                                                                               
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0       985      G   /usr/lib/xorg/Xorg                           214MiB |
|    0      1593      G   compiz                                        95MiB |
|    0      1973      G   fcitx-qimpanel                                 9MiB |
|    0      9093      G   /usr/bin/vlc                                   8MiB |
+-----------------------------------------------------------------------------+

PCI 接口信息

lspci |grep NVIDIA

GeForce GTX 950安装CUDA 9.0后，PCI显卡显示如下

01:00.0 VGA compatible controller: NVIDIA Corporation GM206 [GeForce GTX 950] (rev a1)
01:00.1 Audio device: NVIDIA Corporation Device 0fba (rev a1)