由于CentOS8即将停止维护，所以将系统更换为了Debian，并将TensorFlow在Debian 10上重新安装一下。这里使用的是 Debian 10.11.0版本，机器显卡使用的是1080Ti。如果在CentOS8中安装，可以参考我之前写的步骤： CentOS8安装TensorFlow | OopsDump
先安装Debian 10.11.0，就不按步骤具体说明了。主要注意几点：
1. 安装时，最好选择英文。
2. 分区时，要把默认的系统分区分大一些，或者分出一个 分区 用于挂载/opt目录。
3. 建议安装Gnome桌面环境。
配置Debian
下面的操作都要使用root权限。有一点需要提醒的是，Debian中切换到root用户，要使用“su -”而不能直接用“su”， 不然$PATH中不会包含sbin目录。
关闭自动待机，在界面程序中Settings->Power->Automatic suspend = off。
远程连接的情况，还可以完全禁止休眠和睡眠：（重新开启将mask改为unmask）
# systemctl mask sleep.target suspend.target hibernate.target hybrid-sleep.target
安装完成后，进入系统，配置软件源为本地DVD文件，因为这样可以最快安装，不联网的机器不用担心漏洞问题，也减少自动升级造成的麻烦。先新建个目录 /opt/debian_dvd/ ，将DVD1-3及firmware DVD保存到/opt/debian_dvd/目录下。再在/media/目录创建几个子目录用于开机挂CD，我用了/mdeia/debian_dvd1，/mdeia/debian_dvd2，/mdeia/debian_dvd3，/mdeia/debian_dvd4。然后修改/etc/fstab，增加开机挂CD文件。
(...)
/opt/debian_dvd/debian-10.11.0-amd64-DVD-1.iso /media/debian_dvd1 iso9660 defaults 0 0
/opt/debian_dvd/debian-10.11.0-amd64-DVD-2.iso /media/debian_dvd2 iso9660 defaults 0 0
/opt/debian_dvd/debian-10.11.0-amd64-DVD-3.iso /media/debian_dvd3 iso9660 defaults 0 0
/opt/debian_dvd/firmware-10.11.0-amd64-DVD-1.iso /media/debian_dvd4 iso9660 defaults 0 0
修改/etc/apt/sources.list文件，在其它非注释行首加#改为注释，增加指定DVD为软件源：
deb [trusted=yes] file:/media/debian_dvd1 buster contrib main
deb [trusted=yes] file:/media/debian_dvd2 buster contrib main
deb [trusted=yes] file:/media/debian_dvd3 buster contrib main
deb [trusted=yes] file:/media/debian_dvd4 buster contrib main non-free
deb http://deb.debian.org/debian/ buster main contrib
deb-src http://deb.debian.org/debian/ buster main contrib
如果不想重启机器，可以手动挂载：
# mount -o loop /opt/debian_dvd/debian-10.11.0-amd64-DVD-1.iso /media/debian_dvd1 (其他类似)
更新软件源信息：apt-get update
下面就可以安装一些常用的包了：
# apt-get install vim curl build-essential git python3.7-dev python3-pip linux-headers-$(uname -r) libglvnd-dev dkms pkg-config
将当前用户加入sudo：
# vim /etc/sudoers
(加入一行当前用户)
oopsdump ALL=(ALL) ALL
对于远程的童鞋，要关闭vim的鼠标支持，不能无法用putty上右键粘贴。修改/usr/share/vim/vim81/defaults.vim，将mouse=a该为mouse-=a。
配置远程桌面
安装XRDP远程桌面：
# apt-get install xrdp
# systemctl enable xrdp
如果远程登录后是桌面空白，可以尝试：
# apt-get purge xserver-xorg-legacy
在使用的用户下：（无需root）
# echo gnome-session > ~/.xsession
如果在远程登录后，遇到提示“Authentication is required to create a color profile”，需要输入管理员密码。可以增加下面文件： /etc/polkit-1/localauthority.conf.d/02-allow-colord-oopsdump-com.conf，内容如下：（注意：将下面的{group}替换为你需要允许的用户组， {group} 可以使用video组）
polkit.addRule(function(action, subject) {
if ((action.id == "org.freedesktop.color-manager.create-device" ||
action.id == "org.freedesktop.color-manager.create-profile" ||
action.id == "org.freedesktop.color-manager.delete-device" ||
action.id == "org.freedesktop.color-manager.delete-profile" ||
action.id == "org.freedesktop.color-manager.modify-device" ||
action.id == "org.freedesktop.color-manager.modify-profile") &&
subject.isInGroup("{group}")) {
return polkit.Result.YES;
}
});
要注意的是，如果用户在图形界面没有Logout，远程是无法登录的。可以用SSH执行下面命令，强制用户注销：（DISPLAY值根据实际情况修改，可以ps -ef|grep Xorg查看带“:”的值）
env DISPLAY=:0.0 gnome-session-quit --logout
配置中文环境支持：
# dpkg-reconfigure locales
**选上zh_CN.UTF-8
**默认语言仍为en_US.UTF-8
安装中文输入支持：
# apt-get install fcitx fcitx-googlepinyin
# fcitx-configtool
**在弹出界面点左下+号
**去掉Only Show Current Language的勾
**选中最下的Google Pinyin，然后OK
**然后就可以Ctrl + Space切换输入法
安装VSCode用来编码：
# curl https://packages.microsoft.com/keys/microsoft.asc | gpg --dearmor > microsoft.gpg
# install -o root -g root -m 644 microsoft.gpg /usr/share/keyrings/microsoft-archive-keyring.gpg
# sh -c 'echo "deb [arch=amd64,arm64,armhf signed-by=/usr/share/keyrings/microsoft-archive-keyring.gpg] https://packages.microsoft.com/repos/vscode stable main" > /etc/apt/sources.list.d/vscode.list'
# apt-get install apt-transport-https
# apt-get update
# apt-get install code
安装Nvidia闭源驱动
Nvidia的显卡驱动可以在官方网站下载：https://www.nvidia.com/Download/index.aspx?lang=en-us。选择GeForce类型， GeForce 10系列，GTX 1080Ti。
也可用我的下载地址直接下载：https://us.download.nvidia.cn/XFree86/Linux-x86_64/470.57.02/NVIDIA-Linux-x86_64-470.57.02.run
先关闭原开源的Nouveau 驱动：（原因可以参考早前写的在CentOS8下安装Nviadia驱动的过程。）
# vim /etc/modprobe.d/nouveau
文件中输入：
blacklist nouveau
options nouveau modeset=0
# depmod -ae
# vim /etc/default/grub
文件尾增加：
GRUB_CMDLINE_LINUX_DEFAULT="quiet rd.driver.blacklist=nouveau nouveau.modeset=0"
# update-grub2
# update-initramfs -u
关闭系统GUI，并重启：（使用GUI仍然会自动加载nouveau）
# systemctl set-default multi-user.target
# systemctl reboot
先停止Nouveau的驱动：rmmod nouveau
安装Nvidia的闭源驱动：./NVIDIA-Linux-x86_64-470.57.02.run
恢复图形界面并重启：
# systemctl set-default graphical.target
# systemctl reboot
我发现仍然会加载Nouveau的驱动，但Nvidia的驱动已经可以加载了。这时，可以验证一下：
# nvidia-smi
Thu Nov 1 09:08:34 2021
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 470.57.02 Driver Version: 470.57.02 CUDA Version: 11.4 |
|-------------------------------+----------------------+----------------------+
| 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 ... Off | 00000000:05:00.0 On | N/A |
| 21% 38C P8 11W / 250W | 65MiB / 11176MiB | 0% Default |
| | | N/A |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| 0 N/A N/A 1040 G /usr/lib/xorg/Xorg 39MiB |
| 0 N/A N/A 1070 G /usr/bin/gnome-shell 22MiB |
+-----------------------------------------------------------------------------+
安装CUDA和Cudnn
CUDA 11.4 安装按官网步骤提示即可。 也可用我下载地址： https://developer.download.nvidia.com/compute/cuda/11.4.2/local_installers/cuda_11.4.2_470.57.02_linux.run
安装CUDA 11.4：./cuda_11.4.2_470.57.02_linux.run
安装时，可以不再安装驱动：（因为上面安装的一样版本的驱动）
lqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqk
x CUDA Installer x
x - [ ] Driver x
x [ ] 470.57.02 x
x + [X] CUDA Toolkit 11.4 x
x [X] CUDA Samples 11.4 x
x [X] CUDA Demo Suite 11.4 x
x [X] CUDA Documentation 11.4 x
x Options x
x Install x
x x
x x
x Up/Down: Move | Left/Right: Expand | 'Enter': Select | 'A': Advanced options x
mqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqqj
验证是否安装成功：
# cd /usr/local/cuda/samples/1_Utilities/deviceQuery
# make
.....
# ./deviceQuery
./deviceQuery Starting...
CUDA Device Query (Runtime API) version (CUDART static linking)
Detected 1 CUDA Capable device(s)
Device 0: "NVIDIA GeForce GTX 1080 Ti"
CUDA Driver Version / Runtime Version 11.4 / 11.4
CUDA Capability Major/Minor version number: 6.1
Total amount of global memory: 11176 MBytes (11719409664 bytes)
(028) Multiprocessors, (128) CUDA Cores/MP: 3584 CUDA Cores
GPU Max Clock rate: 1582 MHz (1.58 GHz)
Memory Clock rate: 5505 Mhz
Memory Bus Width: 352-bit
L2 Cache Size: 2883584 bytes
Maximum Texture Dimension Size (x,y,z) 1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
Maximum Layered 1D Texture Size, (num) layers 1D=(32768), 2048 layers
Maximum Layered 2D Texture Size, (num) layers 2D=(32768, 32768), 2048 layers
Total amount of constant memory: 65536 bytes
Total amount of shared memory per block: 49152 bytes
Total shared memory per multiprocessor: 98304 bytes
Total number of registers available per block: 65536
Warp size: 32
Maximum number of threads per multiprocessor: 2048
Maximum number of threads per block: 1024
Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
Max dimension size of a grid size (x,y,z): (2147483647, 65535, 65535)
Maximum memory pitch: 2147483647 bytes
Texture alignment: 512 bytes
Concurrent copy and kernel execution: Yes with 2 copy engine(s)
Run time limit on kernels: Yes
Integrated GPU sharing Host Memory: No
Support host page-locked memory mapping: Yes
Alignment requirement for Surfaces: Yes
Device has ECC support: Disabled
Device supports Unified Addressing (UVA): Yes
Device supports Managed Memory: Yes
Device supports Compute Preemption: Yes
Supports Cooperative Kernel Launch: Yes
Supports MultiDevice Co-op Kernel Launch: Yes
Device PCI Domain ID / Bus ID / location ID: 0 / 5 / 0
Compute Mode:
< Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >
deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 11.4, CUDA Runtime Version = 11.4, NumDevs = 1
Result = PASS
# make clean
...
需要手动下载Cudnn来安装： https://developer.nvidia.com/rdp/cudnn-download
下载Cudnn需要注册，如果不想注册，可以使用我下载用的地址： https://developer.nvidia.com/compute/machine-learning/cudnn/secure/8.2.4/11.4_20210831/cudnn-11.4-linux-x64-v8.2.4.15.tgz
安装方法：
# cd /usr/local/
# tar -xzvf <your save directory>cudnn-11.4-linux-x64-v8.2.4.15.tgz
# sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*
安装虚拟运行环境
安装virtualenv，来进行TensorFlow与CentOS本身包的隔离：
# pip3 install --upgrade virtualenv
创建一个公用虚拟环境：
# mkdir -p /var/venvs/
# virtualenv --system-site-packages /var/venvs/tensorflow
将下面的Diff信息修改到/var/venvs/tensorflow/bin/activate：
unset _OLD_VIRTUAL_PYTHONHOME
fi
+ if ! [ -z "${_OLD_VIRTUAL_LIB:+_}" ] ; then
+ LD_LIBRARY_PATH="$_OLD_VIRTUAL_LIBPATH_OPPSDUMP_COM"
+ export LD_LIBRARY_PATH
+ unset _OLD_VIRTUAL_LIBPATH_OPPSDUMP_COM
+ fi
# This should detect bash and zsh, which have a hash command that must
_OLD_VIRTUAL_PATH="$PATH"
- PATH="$VIRTUAL_ENV/bin:$PATH"
+ PATH="$VIRTUAL_ENV/bin:/usr/local/cuda/bin:$PATH"
export PATH
+ _OLD_VIRTUAL_LIBPATH_OPPSDUMP_COM="$LD_LIBRARY_PATH"
+ LD_LIBRARY_PATH="/usr/local/cuda/lib64:$LD_LIBRARY_PATH"
+ export LD_LIBRARY_PATH
+ export CUDADIR=/usr/local/cuda
# unset PYTHONHOME if set
当需要使用TensorFlow时，先进行：
# source /var/venvs/tensorflow/bin/activate
安装TensorFlow
安装TensorFlow：
**CPU版本
# pip3 install --upgrade tensorflow
**GPU版本
# pip3 install --upgrade tensorflow-gpu
**旧版本
# pip3 install tensorflow=={package_version}
**其他包PyTorch
# pip3 install --upgrade torch8
**如果安装tensorflow时出现h5py安装问题，可以使用下面方法：
# env H5PY_SETUP_REQUIRES=0 pip3 install -U --no-build-isolation h5py==3.1.0
测试命令：
$ python3 -c "import tensorflow as tf;print(tf.reduce_sum(tf.random.normal([1000, 1000])))"
如果遇到问题：
/usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/dtypes.py:516: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
_np_qint8 = np.dtype([("qint8", np.int8, 1)])
/usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/dtypes.py:517: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
_np_quint8 = np.dtype([("quint8", np.uint8, 1)])
/usr/local/lib/python3.7/dist-packages/tensorflow/python/framework/dtypes.py:518: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
(...)
降低numpy版本：
# pip3 uninstall numpy
# pip3 install numpy==1.16.4
再次运行可以成功了。
