Install anaconda+pycharm
Just look at the links below.
Reprinted: https://www.cnblogs.com/zhangkanghui/p/10756630.html
Install cuda+cudnn
1. Check the cuda version corresponding to the graphics card driver version
First open the Control Panel, find "Programs and Features" to check the system graphics card driver version.
The 398.27 in the picture above, and then go here to check the corresponding cuda version, this table.
2. Download and install corresponding CUDA
Go here to download the corresponding cuda, and run it as an administrator after downloading.
Select custom installation, and decide whether to install according to the version of each option. 
Note: Remove the Visual Studio integration option in Customization.
Then test it, enter the dos command in the win terminal
nvcc -V
Note: Here V is uppercase
output
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2018 NVIDIA Corporation
Built on Tue_Jun_12_23:08:12_Central_Daylight_Time_2018
Cuda compilation tools, release 9.2, V9.2.148
Then the installation is successful.
3. Download and install the corresponding CUDNN
To download the library corresponding to cuda on the NVIDIA website , you need to register an account.
After downloading, unzip and copy the .h .lib and .dll files to the corresponding folders in the cuda installation directory.
At this point, the installation should be successful. As for adding the environment, I have not found that it must be added so far. If there is a problem, I will add it later.
4. Download the corresponding version of pytorch
First enter the environment corresponding to anaconda, then go to the pytorch official website to download the version you need, and paste the pytorch download command into the environment corresponding to anaconda to install it.
After the installation is complete, enter python and use the following command to test:
import torch
torch.cuda.is_avaliable()
>>>True # 这里返回true表示gpu版本的pytorch安装成功。