[tensorflow GPU] building of deep learning environment under windows11

I am a Snow fish , a FPGA lovers , The research direction is FPGA Architecture exploration and digital IC Design .
Will be in B Station share IC Relevant teaching courses ,B Stand on your home page ：https://space.bilibili.com/397002941?spm_id_from=333.1007.0.0

QQIC Design &FPGA&DL Communication group Group number ：866169462.

operating system ： Window11 pro
The graphics card ： NVIDIA RTX 3060 Laptop GPU

One 、Anaconda install

I have written the corresponding installation tutorial before ：PyTorch In depth learning notes （ One ）PyTorch Environment configuration and installation https://blog.csdn.net/qq_44447544/article/details/122425322

Two 、CUDA install

NVIDIA Control panel -> System information in the lower left corner -> Components View the supported by the driver CUDA edition ：

Namely support CUDA The maximum version of is 11.6, However, the highest computing platform supported by the flying plasma is CUDA11.2, So I installed 11.2.2 Version of .

CUDA Official download link ：https://developer.nvidia.com/cuda-toolkit-archive

Select operating system related information , Click on 【Download】 Start the download , After downloading, double-click install .

Choose Custom installation , You can choose your own components , If it has been installed VS, The installation of the component can be cancelled , I didn't install it here , So go straight to the next step .

Then select the installation location , Here I am C Plate is 1TB solid state , So I installed it directly according to the default location .

Then you will be prompted to install Visual Studio, Otherwise, some functions cannot be used normally .

3、 ... and 、cudnn install

cuda It can be seen as a workbench , and cudnn Is based on cuda Deep learning acceleration Library , Want to be in cuda For in-depth learning acceleration, you must install cudnn.
Download link on official website ： https://developer.nvidia.com/zh-cn/cudnn

Sign up to log in .

Find a support CUDA11.2.2 Of cuDNN, I'll download it here cuDNN v8.1.1 edition

661MB, Just wait patiently .
After downloading, unzip it , Folder has 3 A folder and a file ：

Copy and paste into the installed CUDA Folder , The default location is C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2, Encounter tips , Select overwrite and replace for all .

Input nvcc -V, Verify that the installation was successful , The information shown in the figure below shows that the installation is successful .

Four 、tensoflow-gpu install

First, select the version to install , open https://tensorflow.google.cn/install/source_windows
Pull to the bottom , You can see gpu Version compatibility information ：

You can see ,CUDA Version is 11.2,cuDNN Version is 8.1 when , Supported by tensorflow_gpu Version is 2.6.0 and 2.5.0, Here we download v2.6.0, then python Version with 3.8.

The specific operation of installation is , open CMD terminal , Enter the following command ：
（1） Creating a virtual environment

conda create -n tf2.6 python=3.8

here tf2.6 For the specified environment name , python=3.8 Specifies the python Version is 3.8.

Input y, then enter ,conda Start downloading the necessary packages automatically . After the download is complete, as shown in the figure below ：

（2） Activate tf2.6 A virtual environment

activate tf2.6

(3) download tensorflow-gpu package

pip install tensorflow-gpu==2.6.0

altogether 423.3MB, Not very much

（4） Verify that the installation was successful .

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

Here I report an error after running , As shown below ：

Could not load dynamic library 'cudart64_110.dll'; dlerror: cudart64_110.dll not found, The lack of cudart64_110.dll file , Go to download . Official download link ：https://www.dll-files.com/cudart64_110.dll.html

Select first Download that will do , Decompress the downloaded compressed package to get cudart64_110.dll , hold cudart64_110.dll Copy and paste to C:/Windows/System32 Li will do :

Re input the output as shown in the following figure , Indicates that... Was successfully installed TensorFlow GPU edition .

Start to be happy and start to learn deeply ！！！