Download Python
(http://python.org/downloads/)
make sure to tick the checkbox “Add Python 3.xx to PATH”

Open “Command Prompt 命令提示字元”, and type in

python --version

(Should pops up >> Python 3.xx <-----version you download)

python -m pip install --upgrade pip

pip install matplotlib

Search your CUDA version for your graphic card
(https://forums.developer.nvidia.com/t/cuda-10-1-and-gtx-1660-ti-not-compatible/79704)
e.g. RTX 1660Ti >>> CUDA toolkit 10.x (but we download newer version CUDA toolkit 11.0)

Check CUDA / Microsoft Visual C++ compatibility
(https://quasar.ugent.be/files/doc/cuda-msvc-compatibility.html)
e.g. CUDA 11.0 >>> Visual C++ 2017

Download Visual Studio
(https://visualstudio.microsoft.com/zh-hant/vs/older-downloads/)

Decide Tensorflow version, and see its cuDNN & CUDA requirments.
(https://www.tensorflow.org/install/source)
e.g. tensorflow-2.5.0 >>> cuDNN 8.1, CUDA 11.2

Go to NVIDIA CUDA site download corresponding CUDA version
(https://developer.nvidia.com/cuda-toolkit-archive)
e.g. CUDA Toolkit 11.2.0

Go to NVIDIA cucDNN site download corresponding cuDNN version
(https://developer.nvidia.com/cudnn)
e.g. cuDNN v8.1.0 (January 26th, 2021), for CUDA 11.0,11.1 and 11.2

Extract the “cudnn" file, and copy “bin, include, lib” folders into local folder,
locate at somewhere like “C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2”.

Search “Edit the system environment variables 檢視進階系統設定”, click right-bottom “”Environment Variables 環境變數..."

Double click “Path” at User variable for xxx.

Copy “bin” and “libnvvp” folder full path, and add into environment virables as new path.
e.g. C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\bin
& C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\libnvvp

Restart computer

After restart, open Command Prompt “run as administrator”

python -m pip install --upgrade pip

pip install tensorflow-gpu

pip3 install --upgrade tensorflow-gpu

Download Visual Studio Code
(https://code.visualstudio.com/download)

Open VScode download extensions Python & Jupyter

test code

import tensorflow as tf
print('tensorflow version', tf.__version)

x = [[3.]]
y = [[4.]]
print('Result: {}'.format(tf.matmul(x,y)))

run with Python 3.10, should come up with

tensorflow version 2.5.0
Result: [[12.]]