之前在本地的Ubuntu机器上安装paddle环境还挺顺利的，但是在多卡服务器上安装确遇到了很多问题，主要是服务器上已经安装了cuda等环境，普通用户也没有权限修改系统的依赖，多卡环境与单卡也有些区别。
主要参考资料就是paddle官方文档paddle安装说明

1. 建立conda虚拟环境

1. 新建虚拟环境

conda create -n paddle_env python=3.9

2. 进入虚拟环境

conda activate paddle_env

2. 安装paddlepaddle gpu版本

这里强调一下，一定要用conda安装，conda安装可以直接在当前环境下安装独立的cuda等依赖，这样就不会与系统预装的依赖环境冲突，pip安装就比较麻烦。

2.1 选择cuda版本

建议在终端输入nvidia-smi 查看系统的CUDA Version，选择比系统版本小的cuda版本，例如我当前机器的CUDA Version是11.5，所以我选择安装cuda 11.2

2.2 安装paddle

conda install paddlepaddle-gpu==2.4.2 cudatoolkit=11.2 -c https://mirrors.tuna.tsinghua.edu.cn/anaconda/cloud/Paddle/ -c conda-forge 

3. 验证及排错

3.1 验证方法

安装完成后您可以使用 python3 进入 python 解释器，输入import paddle ，再输入 paddle.utils.run_check()

如果出现PaddlePaddle is installed successfully!，说明您已成功安装。

3.2 第一次报错：cuda问题

W0505 03:08:12.283917 3969672 dynamic_loader.cc:307] The third-party dynamic library (libcudnn.so) that Paddle depends on is not configured correctly. (error code is /usr/local/cuda/lib64/libcudnn.so: cannot open shared object file: No such file or directory)
  Suggestions:
  1. Check if the third-party dynamic library (e.g. CUDA, CUDNN) is installed correctly and its version is matched with paddlepaddle you installed.
  2. Configure third-party dynamic library environment variables as follows:
  - Linux: set LD_LIBRARY_PATH by `export LD_LIBRARY_PATH=...`
  - Windows: set PATH by `set PATH=XXX;

• 解决方法
  查看环境安装的路径下，其实已经有了cuda相关的依赖：
  
  但是目前还是寻找的系统目录，所以指定到环境目录就可以，在终端输入命令：

export LD_LIBRARY_PATH=[安装路径]/miniconda3/envs/paddle_env/lib

再次验证，可以看到刚才的错误已经不在了。

3.3 第二次报错：NCCL问题（多卡）

W0505 03:22:18.677640 3977430 dynamic_loader.cc:278] You may need to install 'nccl2' from NVIDIA official website: https://developer.nvidia.com/nccl/nccl-downloadbefore install PaddlePaddle.
[2023-05-05 03:22:18,678] [ WARNING] install_check.py:281 - PaddlePaddle meets some problem with 4 GPUs. This may be caused by:
 1. There is not enough GPUs visible on your system
 2. Some GPUs are occupied by other process now
 3. NVIDIA-NCCL2 is not installed correctly on your system. Please follow instruction on https://github.com/NVIDIA/nccl-tests 
 to test your NCCL, or reinstall it following https://docs.nvidia.com/deeplearning/sdk/nccl-install-guide/index.html
[2023-05-05 03:22:18,679] [ WARNING] install_check.py:289 - 
 Original Error is: (PreconditionNotMet) The third-party dynamic library (libnccl.so) that Paddle depends on is not configured correctly. (error code is libnccl.so: cannot open shared object file: No such file or directory)
  Suggestions:
  1. Check if the third-party dynamic library (e.g. CUDA, CUDNN) is installed correctly and its version is matched with paddlepaddle you installed.
  2. Configure third-party dynamic library environment variables as follows:
  - Linux: set LD_LIBRARY_PATH by `export LD_LIBRARY_PATH=...`
  - Windows: set PATH by `set PATH=XXX; (at /paddle/paddle/phi/backends/dynload/dynamic_loader.cc:305)

PaddlePaddle is installed successfully ONLY for single GPU! Let's start deep learning with PaddlePaddle now.

• 解决方法
  下载安装NCCL，这个需要去NVIDIA 官网下载，下载地址。
  
  下载完解压

tar xvf nccl_2.17.1-1+cuda11.0_x86_64.txz

解压后可以直接把库拷贝到环境安装目录下

这时再次验证即可通过！

4. 设置环境变量，可以不用每次设置依赖目录

• 如果要进入paddle环境，需要设置环境变量

export LD_LIBRARY_PATH=[安装路径]/miniconda3/envs/paddle_env/lib

• 可以设置为每次打开终端，自动设置环境变量

vim ~/.bashrc

再最下边输入

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:[环境目录]/miniconda3/envs/paddle_env/lib

保存退出后，重新打开终端即生效。