GPU Ant1裸金属服务器NVIDIA525+CUDA12.0装机和NCCL验证

1. 替换apt源

sudo sed -i "s@http://.*archive.ubuntu.com@http://repo.huaweicloud.cn@g" /etc/apt/sources.list
sudo sed -i "s@http://.*security.ubuntu.com@http://repo.huaweicloud.cn@g" /etc/apt/sources.list
sudo apt update

2. 安装NVIDIA驱动

wget https://us.download.nvidia.com/tesla/525.105.17/NVIDIA-Linux-x86_64-525.105.17.run
chmod +x NVIDIA-Linux-x86_64-525.105.17.run
./NVIDIA-Linux-x86_64-525.105.17.run

3. 安装CUDA

wget https://developer.download.nvidia.com/compute/cuda/12.0.0/local_installers/cuda_12.0.0_525.60.13_linux.run
sudo sh cuda_12.0.0_525.60.13_linux.run

执行安装cuda时，弹出选择界面，要取消选择安装驱动，再执行安装

4.安装nvidia-fabricmanager（必须和驱动版本保持一致）

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/nvidia-fabricmanager-525_525.105.17-1_amd64.deb
sudo dpkg -i ./nvidia-fabricmanager-525_525.105.17-1_amd64.deb
systemctl start nvidia-fabricmanager.service

5. 安装OFED驱动（镜像版本内核较低，只能使用4.9版本驱动）

ofed老版本卸载

cd /usr/sbin
./ofed_uninstall.sh

ofed驱动下载地址：https://network.nvidia.com/products/infiniband-drivers/linux/mlnx_ofed/

wget https://content.mellanox.com/ofed/MLNX_OFED-4.9-7.1.0.0/MLNX_OFED_LINUX-4.9-7.1.0.0-ubuntu20.04-x86_64.tgz
tar -zxvf MLNX_OFED_LINUX-4.9-7.1.0.0-ubuntu20.04-x86_64.tgz
cd MLNX_OFED_LINUX-4.9-7.1.0.0-ubuntu20.04-x86_64
apt-get install -y python3 gcc quilt build-essential bzip2 dh-python pkg-config dh-autoreconf python3-distutils debhelper make
./mlnxofedinstall --add-kernel-support  --force

6. 安装nv-peer-memory

由于OFED驱动版本为4.9，因此只能安装1.1版本nv-peer-memory

下载 https://github.com/Mellanox/nv_peer_memory/tree/1.1-0 压缩包，上传至制作镜像服务器，并解压，执行下面命令进行安装

cd nv_peer_memory-1.1-0/
./build_module.sh
cd /tmp
tar xzf /tmp/nvidia-peer-memory_1.1.orig.tar.gz
cd nvidia-peer-memory-1.1
dpkg-buildpackage -us -uc
dpkg -i ../nvidia-peer-memory-dkms_1.1-0_all.deb

安装完成拷贝配置文件到系统路径中

cp /tmp/nvidia-peer-memory-1.1/nv_peer_mem.conf  /etc/infiniband/
cp /tmp/nvidia-peer-memory-1.1/debian/tmp/etc/init.d/nv_peer_mem   /etc/init.d/

查看nv-peer-memory状态

/etc/init.d/nv_peer_mem/ status

7. 设置openmpi环境变量至bashrc中

先使用如下命令查询openmpi版本号

ls /usr/mpi/gcc/

然后编辑~/.bashrc文件，在末尾行加入如下命令，版本号替换为上述命令查询的版本号

# 加入到~/.bashrc
export LD_LIBRARY_PATH=/usr/local/cuda/lib:usr/local/cuda/lib64:/usr/include/nccl.h:/usr/mpi/gcc/openmpi-4.0.3rc4/lib:$LD_LIBRARY_PATH
export PATH=$PATH:/usr/local/cuda/bin:/usr/mpi/gcc/openmpi-4.0.3rc4/bin

8.  安装docker

curl https://get.docker.com | sh && sudo systemctl --now enable docker

9.  安装nvidia-container-toolkit

配置nvidia-container-toolkit

distribution=$(. /etc/os-release;echo $ID$VERSION_ID) && curl -fsSL https://nvidia.github.io/libnvidia-container/gpgkey | sudo gpg --dearmor -o /usr/share/keyrings/nvidia-container-toolkit-keyring.gpg && curl -s -L https://nvidia.github.io/libnvidia-container/$distribution/libnvidia-container.list | sed 's#deb https://#deb [signed-by=/usr/share/keyrings/nvidia-container-toolkit-keyring.gpg] https://#g' | sudo tee /etc/apt/sources.list.d/nvidia-container-toolkit.list

安装nvidia-container-toolkit

apt-get update
apt-get install -y nvidia-container-toolkit
nvidia-ctk runtime configure --runtime=docker
systemctl restart docker

10. 安装nccl

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/cuda-keyring_1.0-1_all.deb
sudo dpkg -i cuda-keyring_1.0-1_all.deb
sudo apt update
sudo apt install libnccl2=2.16.2-1+cuda12.0 libnccl-dev=2.16.2-1+cuda12.0

11. 安装nccl-test

通过 ls /usr/mpi/gcc/ 查看openmpi的具体版本

cd /root
git clone https://github.com/NVIDIA/nccl-tests.git
cd ./nccl-tests
make  MPI=1 MPI_HOME=/usr/mpi/gcc/openmpi-4.0.3rc4 -j 8

12. 测试验证

a. 验证docker内使用gpu和cuda

拉取pytorch官方镜像

docker run -ti --runtime=nvidia --gpus all pytorch/pytorch:latest bash

在容器中执行nvidia-smi验证gpu

在python中执行如下命令验证cuda

python
import torch
torch.cuda.is_available()

b. nccl测试

单机测试：

/root/nccl-tests/build/all_reduce_perf -b 8 -e 1024M -f 2 -g 8

多机测试：

mpirun --allow-run-as-root --hostfile hostfile -mca btl_tcp_if_include eth0 -mca btl_openib_allow_ib true -x NCCL_DEBUG=INFO -x NCCL_IB_GID_INDEX=3 -x NCCL_IB_TC=128 -x NCCL_ALGO=RING -x NCCL_IB_HCA=^mlx5_bond_0 -x LD_LIBRARY_PATH -x NCCl_IB_QPS_PER_CONNECTION=4 /root/nccl-tests/build/all_reduce_perf -b 8 -e 11g -f 2 -g 8

配置hostfile

#主机私有Ip  单节点进程数
192.168.20.1 slots=1
192.168.20.2 slots=1

配置免密登录：

#一直yes，生成公钥和私钥
ssh-keygen
#输入另一台机器ip地址及密码，配置免密
ssh-copy-id -i ~/.ssh/id_rsa.pub root@192.168.1.1
#测试免密
ssh root@192.168.1.1