Linux配置Tensorflow2 GPU环境

本文是基于系统已经安装了驱动和CUDA的，假如不会安装驱动和CUDA的，可以参考我写的上一篇文章：https://blog.csdn.net/pdc31czy/article/details/136072017?spm=1001.2014.3001.5501

并且本文是基于HPC写的笔记，普通电脑跳过步骤1.

1. 进入GPU节点

[zychen@sms ~]$ ssh gpunode1

2.查看显卡信息（可看可不看）

[zychen@gpunode1 ~]$ nvidia-smi

3.用wget下载anaconda（如果已经安装了Anaconda，直接跳到步骤6）

https://www.anaconda.com/download/
在这skip register并且找到linux下的下载文件，右键复制下载链接

[zychen@gpunode1 ~]$ cd ~
[zychen@gpunode1 ~]$ mkdir tmp
[zychen@gpunode1 ~]$ cd tmp
[zychen@gpunode1 tmp]$ wget https://repo.anaconda.com/archive/Anaconda3-2024.02-1-Linux-x86_64.sh

ERROR: cannot verify repo.anaconda.com’s certificate, issued by ‘/C=US/O=Let’s Encrypt/CN=E1’:
Issued certificate has expired.
To connect to repo.anaconda.com insecurely, use `–no-check-certificate’.

假如上面显示证书过期问题，根据提示重新用wget下载。

[zychen@gpunode1 tmp]$ wget --no-check-certificate https://repo.anaconda.com/archive/Anaconda3-2024.02-1-Linux-x86_64.sh

4.用bash安装anaconda

[zychen@gpunode1 tmp]$ ls
Anaconda3-2024.02-1-Linux-x86_64.sh

[zychen@gpunode1 tmp]$ bash Anaconda3-2024.02-1-Linux-x86_64.sh
然后就一直按enter，按到出现询问咨询yes or no，输入yes。

安装完会询问你是否希望更新你的shell配置文件，以便在启动新的shell会话时自动初始化conda:
Do you wish to update your shell profile to automatically initialize conda?
This will activate conda on startup and change the command prompt when activated.
If you’d prefer that conda’s base environment not be activated on startup,
run the following command when conda is activated:

conda config --set auto_activate_base false

You can undo this by running conda init --reverse $SHELL? [yes|no]

如果你希望conda在启动新的shell会话时自动激活，并且改变命令提示符，你应该输入yes

5.source the .bash-rc file to add Anaconda to your PATH

[zychen@gpunode1 tmp]$ cd ~
[zychen@gpunode1 ~]$ source .bashrc

6.在Anaconda创建TensorFlow2环境并且激活环境

（前提是已经安装了驱动和anaconda，驱动HPC已经有，anaconda安装步骤见3-5步）
(base) [zychen@gpunode1 ~]$ conda create -n tf2 python=3.9.18
[zychen@gpunode1 ~]$ conda activate tf2

7.将 CUDA 的 PATH 和 LD_LIBRARY_PATH 添加到我们的 shell 配置文件中（这里默认之前已经安装了CUDA文件）

(tf2) [zychen@gpunode1 ~]$ vim ~/.bashrc
在文件末尾添加（按insert）下面两行：
export PATH=/usr/local/cuda/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
（按Esc退出insert模式并且输入 ：wq 保存退出）

8.在激活的虚拟环境（tf2）中，执行以下命令安装TensorFlow及其相关依赖包：

(tf2) [zychen@gpunode1 ~]$ conda install -c conda-forge cudatoolkit=11.2.2 cudnn=8.1.0
(tf2) [zychen@gpunode1 ~]$ pip install --upgrade pip
(tf2) [zychen@gpunode1 ~]$ pip install "tensorflow-gpu<2.11"

这里虽然在conda-forge通道里安装了cuda11.2.2，但是实际用的还是cuda11.8。

9.测试

(tf2) [zychen@gpunode1 ~]$ python -c "import tensorflow as tf; print(tf.__version__);print(tf.test.is_gpu_available())"
或
(tf2) [zychen@gpunode1 ~]$ python -c "import tensorflow as tf; print(tf.config.experimental.list_physical_devices('GPU'))"

10. 假如同时有PyTorch环境和TensorFlow2环境，测试两个环境是否都能够识别到GPU

[zychen@sms ~]$ ssh gpunode1
[zychen@gpunode1 ~]$ conda activate torch39env
(torch39env) [zychen@gpunode1 ~]$ python -c "import torch; print(torch.cuda.is_available()); print(torch.version.cuda)"
(torch39env) [zychen@gpunode1 ~]$ conda deactivate
[zychen@gpunode1 ~]$ conda activate tf2
(tf2) [zychen@gpunode1 ~]$ python -c "import tensorflow as tf; print(tf.config.experimental.list_physical_devices('GPU'))"