CUDA 10.2

CUDA cuDNN Installation

Installing CUDA, cuDNN on Windows 10

This covers the installation of CUDA, cuDNN on Windows 10. This article below assumes that you have a CUDA-compatible GPU already installed on your PC.

See here for more detailed instruction by NVIDIA

Method 1: Installation using Anaconda

Install CUDA and cuDNN with conda in Anaconda prompt.

Here, it is assumed you have already installed Anaconda. If you do not have Anaconda installed, follow How to Install Anaconda

Recommend to use conda virtual environment for specific CUDA version contained in the env.

Installing CUDA =10.2

First, Run Anaconda Prompt(admistration)

(Option1: install in base)

If you want to install the same CUDA version for all environment, install in (base)

conda install -c anaconda cudatoolkit==10.2.89

(Option2: install in Specific Environment)

It is recommended to install specific CUDA version in the selected Python environment.

[$ENV_NAME] is your environment name. e.g. conda activate py37

#conda activate [$ENV_NAME]
conda activate py37
conda install -c anaconda cudatoolkit==10.2.89

Installing cuDNN

Conda will find the compatible cuDNN for the installed version of CUDA

conda install -c anaconda cudnn

The available cuDNN version could be a low version (e.g. cuDNN 7.6.5). If you want to install a higher version, then follow Method 2: Using NVDIA downloader.

image

Method 2: Installation using NVIDIA downloader

Step 1: Check the software you need to install

Assuming that Windows is already installed on your PC, the additional bits of software you will install as part of these steps are:

• Microsoft Visual Studio Community(v2017 or higher )

• NVIDIA CUDA Toolkit

• NVIDIA cuDNN

Step 2: Install Visual Studio Community

Visual Studio is a Prerequisite for CUDA Toolkit

Visual Studio Community is required for the installation of Nvidia CUDA Toolkit. If you attempt to download and install CUDA Toolkit for Windows without having first installed Visual Studio, you get a message for installation.

Follow: How to install Visual Studio Community

Step 3: Install CUDA Toolkit for Windows 10

(updated 2022.2 : Install CUDA Toolkit 10.2 )

For more detailed instructions, see Nvidia CUDA installation guide for windows.

Installing CUDA Toolkit 10.2

• Go to CUDA download center

  • The CUDA Toolkit (free) can be downloaded from the Nvidia website. Click here to go to the download center

NVIDIA CUDA Toolkit 10.2 DownloadNVIDIA Developer

You should check which version of CUDA Toolkit you choose for download and installation to ensure compatibility with Pytorch or Tensorflow

For Latest CUDA Toolkit : check here

• Select Window10 , exe(Network). Download the Base Installer and Run.

• After Downloading, Install(Recommended Option). It should take 10~20min to install.

  • It will be installed in C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2

Step 4: Install CUDA patches

After CUDA installation, install additional Patches for CUDA Toolkit.

Step 5: Install cuDNN

After installing CUDA and its patches, the next step is to find a compatible version of CuDNN.

• Check which version of **** cuDNN is needed for specific Tensorflow or Pytorch

---

Step 5.1: Register NVIDIA

Visit CuDNN website to download.

First, you have to register to become a member of the NVIDIA Developer Program (free).

Step 5.2: Install cuDNN 8.0.5 for CUDA 10.2

Select Download cuDNN.

image

Select cuDNN version for CUDA 10.2. You can also check cuDNN Archive if you cannot find the version.

cuDNN v8.0.5 for CUDA 10.2 or

cuDNN v8.3.3 for CUDA 10.2

image

1. Unzip the cuDNN package.

   cudnn-windows-x86_64-*-archive.zip

2. Copy the following files from the unzipped package into the NVIDIA CUDA directory.

   • CUDA directory: C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.2

   Do not replace the whole folders. Just copy the files.

3. Check the installed CUDA in your computer. Run the Command Prompt and type

nvcc --version