Jetson nano 安裝Tensorflow-GPU CUDA連結

此篇簡單介紹如何在nano上安裝python3的tensorflow gpu版

事先安裝所需要的套件:

sudo apt-get install libhdf5-serial-dev hdf5-tools

sudo apt-get install python3-pip

sudo pip3 install -U pip

sudo apt-get install zlib1g-dev zip libjpeg8-dev libhdf5-dev

sudo pip3 install -U numpy grpcio absl-py py-cpuinfo psutil portpicker grpcio six mock requests gast h5py astor termcolor

接下來安裝tensorflow gpu 版:

sudo pip3 install --pre --extra-index-url https://developer.download.nvidia.com/compute/redist/jp/v42 tensorflow-gpu

或者 安裝自己想要的tensorflow gpu的版本指令範例如下（tensorflow-gpu=1.13.1+nv19.3）

pip3 install --extra-index-url https://developer.download.nvidia.com/compute/redist/jp/v42 tensorflow-gpu==1.13.1+nv19.3 --user

安裝完後，tensorflow-gpu記得要加入CUDA連結，才能正常使用tensorflow-gpu

(下指令nvcc -V是不會成功的，會顯示沒有相對應的指令)

加入CUDA連結:

下指令打開bashrc

sudo gedit  ~/.bashrc

在最後面加入這三行後儲存並關閉（記得關閉後需重新登入才會生效）

export CUBA_HOME=/usr/local/cuda-10.0

export LD_LIBRARY_PATH=/usr/local/cuda-10.0/lib64:$LD_LIBRARY_PATH

export PATH=/usr/local/cuda-10.0/bin:$PATH

後記:

嘗試在python3上跑人臉辨識(from: davidsandberg/facenet)

基本上連開啟都有困難，會顯示程式佔用記憶體過大，系統能用的記憶體過低之類的資訊。

參考 reference:

安裝tensorflow-gpu:

https://medium.com/aiot-taipei/jeston-nano-tensorflow-gpu-%E5%AE%89%E8%A3%9D-b26d42f7c3f3

https://blog.csdn.net/beckhans/article/details/89146881

CUDA連結:

https://blog.csdn.net/deeplearningfeng/article/details/89554181