ubuntu16.04 安装horovod

1.下载安装nccl

1.1 下载仓库安装包后

sudo dpkg -i /home/chen/mydisk/YJY/nccl-repo-ubuntu1604-2.8.3-ga-cuda10.1_1-1_amd64.deb

1.2 更新: sudo apt update

1.3 安装nccl

sudo apt install libnccl2=2.8.3-1+cuda10.1 libnccl-dev=2.8.3-1+cuda10.1

1.4 测试nccl是否安装好

git clone https://github.com/NVIDIA/nccl-tests.git
cd nccl-tests
make
./build/all_reduce_perf -b 8 -e 256M -f 2 -g 'ngpus'

2.下载openmpi

2.1 下载安装包后

 tar -xvzf /home/chen/mydisk/YJY/openmpi-4.0.5.tar.gz 
cd openmpi-x.x.x

./configure --prefix="/usr/local/openmpi"
make -j 8
sudo make install

2.2 在.bashrc中添加环境变量

sudo gedit ~/.bashrc
export PATH="$PATH:/usr/local/cuda/bin:/usr/local/openmpi/bin"
export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:/usr/local/cuda/lib64:/usr/local/openmpi/lib/"
source ~/.bashrc

打开新的终端使之生效。

3.安装cmake

sudo apt install cmake

3.1 查看cmake版本

cmake --version

4.下载horovod

4.1 安装好nccl之后可使用以下命令安装horovod

HOROVOD_GPU_OPERATIONS=NCCL pip install horovod -i https://pypi.tuna.tsinghua.edu.cn/simple