SoFunction
Updated on 2024-12-19

Graphic tutorial for PyTorch environment installation

1. Install CUDA

Installer download address (also mentioned in the main blog when describing version selection)

Official website of various versions of CUDA download address
Official website various cuDNN download address

Open "cuda_8.0.44_win10.exe", this process will be very slow, be patient (this also suggests that I should change my computer)

Choose the unzip address (it's temporary anyway, just the C drive, not a big problem)

Start decompression (this process is still very slow, especially my computer at that time do not know what the problem of CPU and memory usage is unusually high, this process is also a lot of time I spent)

After decompression, add in the installer

Start Installation

4. Knock on the blackboard! Don't choose the default Lite here, it should be changed to All (see the small print below to indicate that this is Family Bucket), not that installing Family Bucket isn't possible, but there's one thing about installing it that will always cause the installation to fail

Especially this visual studio integration should never be chosen!

It is sufficient to choose the following installation

Just install it on the C drive, it's easier to path it somewhere else later.

Successful installation

Verify successful installation

(1) Environment variables should have been added automatically

(2) View version information in cmdnvcc -V

(3) View the GPU runtime monitoring interface after entering under the path

(4) Run (you need to go to the directory first)

(5) Run (you need to go to the directory first)

2. Install cuDNN

Official website various cuDNN download address
cuDNN can't be called installation, you just need to unzip the downloaded package and put the corresponding folder of files into the corresponding folder under the CUDA installation path.

3. Uninstallation

cuDNN is supposed to just copy the files into the CUDA installation directory, so just delete it (deleting the whole folder directly after uninstalling CUDA is also fine)
CUDA uninstallation: Control Panel - uninstall the program (do not use 360 and other antivirus software, can not find the corresponding program), in accordance with the installation time sorting, the top of these with the version number, is just installed CUDA, uninstall one by one to be able to

Installation Reference:

1.video tutorial: 3 minutes deep learning [environment building] CUDA +Anaconda simple rough_Beep_Meep]

2. windows10 under CUDA11.1, cuDNN8.0, tensorflow-gpu2.4.1 installation tutorials and problem solving method _

3.Installation and uninstallation of CUDA, cuDNN in Win10

Detailed Installation - Strongly Recommended!

This article on the PyTorch environment to install the graphic tutorials on the article is introduced to this, more related to the installation of the PyTorch environment, please search for my previous articles or continue to browse the following related articles I hope you will support me in the future!