paddle2.3+centos7.6.5+cuda10.1+nccl2.7.8

查看适配确定cuda版本之后，注册并登录英伟达官网，查找指定cuda版本的NCCL软件

cuda cudnn安装

下载路径：https://developer.nvidia.com/nccl/nccl-legacy-downloads

1. https://developer.nvidia.com/compute/machine-learning/nccl/secure/v2.7/prod/nccl-repo-rhel7-2.7.8-ga-cuda10.1-1-1.x86_64.rpm

2. https://developer.nvidia.com/compute/machine-learning/nccl/secure/v2.7/prod/agnostic/x64/nccl_2.7.8-1+cuda10.1_x86_64.txz 。依赖包，下载好后，需tar -xvf a.txz解压

tar -xvf nccl_2.7.8-1+cuda10.1_x86_64.txz  #包含 lib, include
sudo cp include/* /usr/local/cuda/include
sudo cp -r lib/* /usr/local/cuda/lib64
cd /usr/local/cuda/lib64
sudo rm libnccl.so libnccl.so.2              # 删除原有文件
sudo ln -s libnccl.so.2.7.8 libnccl.so.2  #  创建软连接
sudo ln -s libnccl.so.2 libnccl.so           #  创建软连接
ls # 查看软链接是否创建成功
echo "export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64" >> ~/.bashrc
1
2
3
4
5
6
7
8
9

验证

>>> import paddle
>>> paddle.utils.run_check()
Running verify PaddlePaddle program ...
PaddlePaddle works well on 8 GPUs.
PaddlePaddle is installed successfully! Let's start deep learning with PaddlePaddle now.

1
2
3
4
5
6