Win10 uses conda to configure tensorflow-gpu environment

I still want to record the environment once. I am really exhausted. I hope that I will configure the environment less and use my brain to read and change the code.

1. Conda source change

It is too slow to directly create a virtual environment, and you need to change to a domestic source.
Tsinghua University open source mirror station Anaconda mirror usage help

2. Create a virtual environment

conda create -n tfgpu python=3.9

3. Install related packages

pip install --ignore-installed --upgrade tensorflow-gpu
conda install cudatoolkit=11.2 cudnn -c conda-forge

pip install scikit-learn
pip install numpy matplotlib pandas
pip install ipykernel --upgrade

Four, cuda problem

Program running problem: Could not load dynamic library 'cudart64_110.dll'; dlerror: cudart64_110.dll not found solution

File download URL,
just download one,
unzip the dll and put it in C:\Windows\System32\

5. Others

When encountering a problem, solve it, don't put it badly, work hard to solve it, and use your brain more.