There is no need to install CUDA and cudnn manually. You can install tensorflow GPU through a one-line program. Take tensorflow gpu2.0.0, cuda10.0, cudnn7.6.5 as examples

stay anaconda prompt In program base In the environment ：

# Create a tf2 Virtual environment for , And according to the preset environment name tensorflow-gpu

# Automatic installation CUDA,cuDNN,TensorFlow GPU etc.

conda create -n tf2 tensorflow-gpu==2.0.0

# Activate tf2 A virtual environment conda activate tf2

conda activate tf2

This quick installation method is called minimalist installation method . This is also used Anaconda The convenience of distribution The place of . Installed via minimalist TensorFlow, When using, you need to activate the corresponding virtual environment , This needs to be connected with Standard Edition differentiation . The standard version is installed on Anaconda The default environment for base in , Generally, manual activation is not required base Ring habitat .

# Use Tsinghua source to install common python library

pip install -U ipython numpy matplotlib pillow pandas -i https://pypi.tuna.tsinghua.edu.cn/simple

TensorFlow At run time , By default, it will occupy all GPU Video memory resources , This is very unfriendly behavior , especially When the computer has multiple users or programs at the same time GPU Resource time , Take all of GPU Video memory resources will make Other programs cannot run . therefore , General recommended settings TensorFlow The video memory occupation mode of is the increasing occupation mode , That is, apply for video memory resources according to the actual model size , The code implementation is as follows ：

gpus = tf.config.experimental.list_physical_devices('GPU')
if gpus:
    try:
    #  Set up  GPU  Growth occupancy 
        for gpu in gpus:
            tf.config.experimental.set_memory_growth(gpu, True)
    except RuntimeError as e:
    #  Printing exception 
        print(e)