Install anaconda+pytorch+CUDA10.2+cuDDN in windows10 environment

anaconda is not to say, the official website to download and install
Download: https://www.anaconda.com/download/
.
.
.
Step One: Install CUDA

Step 2: Install cuDDN

Step three: Test
.
.
.

One: Install CUDA

NVIDIA Control Panel→System Information->Components, check the version of NVCUDA.DLL
Insert picture description here
Insert picture description here
and then enter the CUDA official website to download the corresponding version of CUDA:
https://developer.nvidia.com/cuda-downloads?target_os=Windows&target_arch=x86_64&target_version=10&target_type=exelocal
( Note that after querying the local CUDA version, you can install the lower version of CUDA. My computer query is CUDA11.0, and then I installed the 10.2 version of CUDA)
Insert picture description here
You can choose to customize the installation, all the way is the default.
Insert picture description here
Insert picture description here
After the installation is complete (my own process is not saved, the picture is from the network),
start the cmd command line, and then enter:

cd C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2\bin

(I installed version 10.2)
Then enter:

nvcc -V 

Note that the capital "V", the installation is successful as shown in the figure below.
Insert picture description here
.
.
.

Two: Install cuDDN

Download cuDDN URL:
https://developer.nvidia.com/rdp/cudnn-download
need to register a NVIDIA account, and then select the following diagram:
Insert picture description here
Once downloaded, unzip:
Insert picture description here
and these three files under the folder Insert picture description here
under the file separately Copy it to the corresponding bin, include, and lib directories in the C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2 directory.
Finally, open "this computer" → "advanced system settings" → "environment variables" → "system variables" → "path" to check whether the bin directory exists, if not, add it.
.
.
.

Three: Test:

Insert picture description here
Same as my output, that's it!
.
.
.

If

Here your output is wrong. Check if your pytorch is without CUDA version. You can delete pytorch and reinstall it:
delete command:

pip uninstall torch

Then enter the pytorch official website to select the appropriate version:
https://pytorch.org/get-started/locally/

It is recommended to choose the pip method to install. The conda method reported messy errors when I installed it. The command I ran directly in the terminal in pycharm at that time, the running command was:

pip install torch===1.5.0 torchvision===0.6.0 -f https://download.pytorch.org/whl/torch_stable.html

About an hour and a half, the installation is complete:
Insert picture description here
ok, over, I hope I can help everyone~

Guess you like

Origin blog.csdn.net/Su_Del/article/details/106530652