1.创建conda环境 conda create -n tf2 python=3.7

2.进入conda环境 conda activate tf2

3.输入 nvidia-smi 查看有没有显卡驱动。(没有安一个，不管是windows/linux)

4. 安装cudatoolkit 和 cuDNN

                conda install cudatoolkit=10.0 cudnn

5. 安装tensorflow

                pip install tensorflow-gpu==2.0.0 -i https://pypi.tuna.tsinghua.edu.cn/simple

6. 验证安装

                python -c "import tensorflow as tf; print(tf.test.is_gpu_available())"

                显示‘True’就是没毛病。

注：可能出现问题：

TypeError: Descriptors cannot not be created directly. If this call came from a _pb2.py file, your generated code is out of date and must be regenerated with protoc >= 3.19.0. If you cannot immediately regenerate your protos, some other possible workarounds are:
1. Downgrade the protobuf package to 3.20.x or lower.
2. Set PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python (but this will use pure-Python parsing and will be much slower).

方法：

                pip uninstall protobuf

                pip install protobuf==3.20.1

另一类最常见的是版本不对应。注意cudatoolkit，cudnn，tensorflow-gpu的版本对应关系。如图

具体可查看tensorflow官网