centos7安装cuda步骤 RTX2060

参考https://www.jianshu.com/p/1710d9b58a50

增加新yum源
yum install epel-release

安装内核
yum install kernel-devel-$(uname -r) kernel-headers-$(uname -r)

安装GCC9
yum install centos-release-scl scl-utils-build -y &&
yum install devtoolset-9-toolchain -y &&
scl enable devtoolset-9 bash&& echo “source /opt/rh/devtoolset-9/enable” >>/etc/profile
#yum list all --enablerepo=‘centos-sclo-rh’ | grep “devtoolset-”

安装lspci工具
yum install pciutils #lspci | grep -i nvidia

#安装dkms（前面有新源，可以不手动安装）
#yum install dkms

下载cuda
wget https://developer.download.nvidia.com/compute/cuda/11.2.2/local_installers/cuda-repo-rhel7-11-2-local-11.2.2_460.32.03-1.x86_64.rpm

安装cuda
sudo rpm -i cuda-repo-rhel7-11-2-local-11.2.2_460.32.03-1.x86_64.rpm

安装
sudo yum clean all
sudo yum -y install nvidia-driver-latest-dkms cuda
sudo yum -y install cuda-drivers

环境变量设置
export PATH=/usr/local/cuda-11.2/bin:KaTeX parse error: Expected 'EOF', got '&' at position 8: {PATH} &̲& export LD_LIB…{LD_LIBRARY_PATH}

echo “export PATH=/usr/local/cuda-11.2/bin:${PATH}”$‘\n’“export LD_LIBRARY_PATH=/usr/local/cuda-11.2/lib64:${LD_LIBRARY_PATH}” >> /etc/profile

查看nvdia驱动
nvidia-smi
nvcc -V

安装cudnn
tar -xf *.tar.xz

将cudnn的内容复制到cuda安装后对应目录即可

关于安装Tensorflow

注意python版本和tensorflow版本，和cuda对应

提示问题： I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
解决方法：
for a in /sys/bus/pci/devices/*; do echo 0 | sudo tee -a $a/numa_node; done