Anaconda

Anaconda Is to manage each python Package tools , Here we mainly use numpy, And some other commonly used bags .

Anaconda The website links ：https://www.anaconda.com/

Of each system version Anaconda The installer can be downloaded and installed directly , For relevant usage methods, please refer to The article

CUDA

1. First Check your computer GPU Corresponding cuda edition , As shown in the figure open NVIDIA Control panel -> help -> system information -> Components （ It can be opened in the control center or the notice bar in the lower right corner ）, I'm here GPU Corresponding cuda The version is 11.6.106, So the installation cuda Cannot exceed this version .

Or in cmd Command line input nvidia-smi You can also view CUDA edition .

2. Get into CUDA Toolkit Archive Website Select a version lower than the one just found cudatoolkit, This article chooses CUDA Toolkit 11.0 Update 3, Download installation package about 2.7G.

Be careful ： Here's the choice cuda The version will decide to install later cudnn and tensorflow-gpu Version of , The three are related , If the version does not match, you will not be able to use , Specific matching version reference https://tensorflow.google.cn/install/source_windows, The following table is an excerpt .

3. Double-click the downloaded exe Installation package , choice Temporary decompression Location , And then click ok, The decompression time is about two minutes .

4. Agree to the license agreement , And then choose Custom installation Options , And then click next step

5. Select driver components , must do Check CUDA, And then click next step .

6. Choose an installation location on your computer , then Start installation .

7. Wait for the installation to finish , This computer 18 Notebook mechanical hard disk installation It takes about 5 minute

8. Check environment variables , Generally, the environment variables will be automatically configured after the installation , open Set up - System - system information - Advanced system setup - environment variable Inspection , If not, you need to add it yourself （ Note that when adding, change to your own installation path ）

• Whether there is CUDA and NVCUDASAMPLES Two sets of environment variables .

• Open Path, See if there are the following environment variables .

9. CUDA installation is complete

cnDNN

1. open cudnn Official website , Check what you should download according to the table above cuDNN edition , my CUDA Version is 11.0, Corresponding cuDNN Version is 8.0, Pay attention to see clearly cuDNN There is a corresponding CUDA edition , One cuDNN It may correspond to multiple versions CUDA, When downloading, you will be required to register first NVIDIA account number .

2. Unzip the file , Move the unzipped file to CUDA In the corresponding installation directory

3. cuDNN installation is complete

install TensorFlow

1. open cmd Command line interface （ It is recommended to open as an administrator ）, Switch opentuna Of pip Mirror source

pip config set global.index-url https://opentuna.cn/pypi/web/simple

2. Creating a virtual environment , Followed by the computer installation python edition , Pay attention to whether it is related to

conda create -n tensorflow2 python=3.8

3. Activate the virtual environment

conda activate tensorflow2

4. install tensorflow, The specific version number corresponds to the table above .

pip install tensorflow-gpu==2.3.0

5. Verify that the installation was successful , open cmd Interface , Enter the following command , Check whether the output Successfly

python
import tensorflow as tf
tf.config.list_physical_devices('GPU')

If appear Could not load dynamic library 'cudart64_110.dll' Don't panic , It means that the computer is not restarted after the installation , Just restart and run again .

Keras

1. TensorFlow And Keras Refer to the following table for the corresponding version

2. Still running Anaconda Of tensorflow2 Run the following instructions in the environment

pip install keras==2.4.3

This blog post was first published on the personal blog website ：https://www.mahaofei.com/, Welcome to .