tensorflow cuda nivdia
NVIDIA 驱动程序下载
驱动确定cuda tookit版本 https://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html
cuda tookit 版本确定cuDNN版本
CUDA Toolkit Download
Ubuntu 18.10(nvdia 418.56->cuda toolkit 10.1->cuDNN 7.5.0)
(nvidoa 390->cuda8.0->cuDNN6.0)
cuDNN Archive
rope for spyder3
tensorflow与cuda cuDNN版本对应关系(在网页最下部)
cuda9=>1.5~1.12(XLA)
cuda8=>1.0~1.4
1.13.1=>cuda10.0=>cuDNN7.5
tf1.4-gpu=>cuda8.0=>cuDNN6
tensorflow-gpu 1.4.0
配置cuda toolkit环境变量
(a)vim ~/.bashrc
(b)添加如下行,用于添加cuda bin的路径到环境变量PATH
export PATH=$PATH:/usr/local/cuda/bin
(c)添加如下行,用于添加cuda/lib和cuda/lib64路径到环境变量LD_LIBRARY_PATH
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib:/usr/local/cuda/lib64
(d)使配置生效
source ~/.bashrc