Cuda windows how to install

How to Install CUDA on Windows 11

CUDA (Compute Unified Device Architecture) is a parallel computing platform and application programming interface (API) model created by NVIDIA. It allows developers to utilize the power of NVIDIA GPUs for general-purpose processing—often referred to as GPGPU (General-Purpose computation on Graphics Processing Units). CUDA streams are widely used in machine learning, scientific computing, and graphics rendering. If you’re looking to harness the capability of your NVIDIA GPU on Windows 11, this guide will help you through the installation process step-by-step.

1. System Requirements

Before beginning the installation process, ensure that your Windows 11 system meets the following requirements:

• Supported NVIDIA GPU: Visit the official NVIDIA CUDA GPUs list to check if your graphics card is compatible.
• NVIDIA Drivers: You must have the latest NVIDIA drivers installed. This can typically be done through the GeForce Experience application or by downloading the latest drivers directly from the NVIDIA website.
• Windows 10 or 11: CUDA installation is supported on the latest versions of Windows.
• Visual Studio: If you plan to develop CUDA applications, installing Visual Studio (Community, Professional, or Enterprise) is highly recommended as it provides robust development tools.

2. Preparing for Installation

Before installing CUDA, you should prepare your system and download the necessary software.

1. Update Windows: Ensure that Windows 11 is fully updated through Windows Update.
2. Install the Latest NVIDIA Driver:
   • Go to the NVIDIA Driver Download page.
   • Select your graphics card model and download the latest driver.
   • Once downloaded, run the installer and follow the on-screen instructions.
3. Download CUDA Toolkit:
   • Visit the CUDA Toolkit Download page.
   • Choose «Windows» as your operating system and select «x86_64» for your architecture.
   • Select the installer type (typically, the local installer is recommended).
   • Click «Download» to save the installer to your computer.

3. Installing CUDA Toolkit

Once you have everything prepared, you can begin the installation of the CUDA Toolkit.

1. Run the CUDA Installer:

   • Navigate to the directory where you downloaded the CUDA Toolkit.
   • Double-click the installer executable (e.g., cuda_11.6.0_511.79_win.exe).

2. Select Installation Options:

   • When prompted, choose the installation options. You can opt for «Express» (recommended for most users) or «Custom» (if you need specific versions or components).
   • If you select «Custom», be sure to select the components you require, such as:
     • CUDA Toolkit
     • CUDA Samples
     • NVIDIA Driver (if you haven’t installed it earlier)

3. Complete Installation:

   • Follow through the prompts. You may need to accept the End User License Agreement (EULA).
   • Once the installation is complete, you might be prompted to reboot your system. It’s a good idea to do so to ensure all changes take effect.

4. Verifying CUDA Installation

After installation, it’s important to verify that CUDA has been installed correctly.

1. Open Command Prompt:

   • Press Windows + R, type cmd, and hit Enter.

2. Check CUDA Version:

   • In the Command Prompt, type the following command and press Enter:

     nvcc --version

   • This command checks the version of the NVIDIA CUDA compiler driver. You should see output with the version number of CUDA installed.

3. Run Sample Programs:

   • If you opted to install the CUDA Samples, navigate to the installation directory (commonly C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6samples).
   • Open the Command Prompt again and navigate to the samples folder:

     cd "C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6samples"

   • Use the following commands to build and run the CUDA samples:

     cd 0_Simple
     mkdir build
     cd build
     cmake ..
     cmake --build . 

   • After the build completes successfully, run the sample executable (e.g., simpleQuadd or vectorAdd), and check the output.

5. Setting Environment Variables

For CUDA to work correctly with your applications, you need to set the environment variables.

1. Open Environment Variables:

   • Right-click on the Start button and select «System.»
   • Click on «Advanced system settings» from the left pane.
   • In the System Properties window, click on the «Environment Variables» button.

2. Add CUDA to PATH:

   • In the «System variables» section, find the Path variable and click «Edit.»
   • Click «New» and add the following paths (adjust according to your specific version):

     C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6bin
     C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6libnvvp

   • Click «OK» to close all windows.

3. Set CUDA_HOME Variable (optional but recommended):

   • In the Environment Variables window, under «System variables,» click «New.»
   • In the «Variable name» field, enter CUDA_HOME.
   • In the «Variable value» field, enter:

     C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6

   • Click «OK» to save the variable.

6. Installing cuDNN (optional)

For many deep learning applications, you’ll also need the cuDNN library, which provides optimized implementations of standard routines such as convolution, pooling, normalization, and activation layers.

1. Create a NVIDIA Developer Account:

   • Go to the NVIDIA Developer website and create an account if you don’t have one.

2. Download cuDNN:

   • Once logged in, navigate to the cuDNN download section and choose the version that matches your installed CUDA Toolkit.
   • Download the cuDNN library for Windows.

3. Install cuDNN:

   • After extracting the downloaded cuDNN files, copy the contents from the bin, include, and lib folders to their corresponding paths in the CUDA Toolkit directory:
     • bin → C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6bin
     • include → C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6include
     • lib → C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6libx64

4. Verify cuDNN Installation:

   • You can confirm that cuDNN is working by looking for cudnn.h in C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6include and cudnn.lib in C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6libx64.

7. Developing Your First CUDA Application

To create and run a simple CUDA application, you may follow these steps:

1. Open Visual Studio:

   • Start Visual Studio and create a new C++ project.

2. Set CUDA Project Settings:

   • Right-click on the project, go to «Properties.»
   • Under Configuration Properties, ensure that:
     • C/C++ → General → Additional Include Directories includes the CUDA include path:

       C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6include

     • Linker → General → Additional Library Directories includes:

       C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.6libx64

3. Add CUDA Code:

   • In your project, add a new .cu file (CUDA C++ file) and write a simple CUDA kernel:

     #include 
     
     __global__ void vectorAdd(const float* A, const float* B, float* C, int N) {
      int i = threadIdx.x + blockIdx.x * blockDim.x;
      if (i < N) {
          C[i] = A[i] + B[i];
      }
     }
     
     int main() {
      const int N = 1 << 20; // 1 Million elements
      size_t size = N * sizeof(float);
     
      float *A, *B, *C;
      cudaMallocManaged(&A, size);
      cudaMallocManaged(&B, size);
      cudaMallocManaged(&C, size);
     
      // Initialize A and B
      for (int i = 0; i < N; i++) {
          A[i] = static_cast(i);
          B[i] = static_cast(i);
      }
     
      // Launch kernel
      vectorAdd<<>>(A, B, C, N);
     
      // Wait for GPU to finish before accessing on host
      cudaDeviceSynchronize();
     
      // Check for errors
      for (int i = 0; i < N; i++) {
          if (C[i] != A[i] + B[i]) {
              std::cout << "Error: " << C[i] << " != " << A[i] + B[i] << std::endl;
              return 1;
          }
      }
      std::cout << "Success!" << std::endl;
     
      // Free memory
      cudaFree(A);
      cudaFree(B);
      cudaFree(C);
     
      return 0;
     }

4. Build and Run the Application:

   • Build the project, and if successful, run it. You should see Success! printed if everything is set up correctly.

8. Troubleshooting Common Issues

While installing CUDA can be relatively straightforward, you may encounter some issues:

1. Installation Failures: If the installation fails, ensure that:

   • All previous versions of the NVIDIA driver and CUDA are completely uninstalled.
   • You have sufficient permissions (administrator rights) for the installation process.

2. Compatibility Problems: Always ensure that your GPU supports the version of CUDA you’re trying to install by cross-referencing with official NVIDIA documentation.

3. Environment Variable Issues: If nvcc --version does not work, double-check that your environment variables are set correctly and that you’ve restarted your command prompt after making changes.

4. Driver Issues: If your CUDA code does not run as expected, ensure that your drivers are up-to-date. If you’ve installed new drivers, restart your PC.

5. Running GPU Programs: Occasionally, programs may fail to detect or utilize the GPU. Make sure that the CUDA application/device is compatible with your graphics card.

Conclusion

Congratulations! You have successfully installed CUDA on your Windows 11 system. You can now leverage the power of your NVIDIA GPU for parallel computing tasks in machine learning and high-performance computing domains. As you experiment and develop CUDA applications, you’ll unlock greater computational capabilities and efficiencies. Always keep your drivers, CUDA Toolkit, and cuDNN updated to take full advantage of the latest features and improvements.

В очередной раз после переустановки Windows осознал, что надо накатить драйвера, CUDA, cuDNN, Tensorflow/Keras для обучения нейронных сетей.

Каждый раз для меня это оказывается несложной, но времязатратной операцией: найти подходящую комбинацию Tensorflow/Keras, CUDA, cuDNN и Python несложно, но вспоминаю про эти зависимости только в тот момент, когда при импорте Tensorflow вижу, что видеокарта не обнаружена и начинаю поиск нужной страницы в документации Tensorflow.

В этот раз ситуация немного усложнилась. Помимо установки Tensorflow мне потребовалось установить PyTorch. Со своими зависимостями и поддерживаемыми версиями Python, CUDA и cuDNN.

По итогам нескольких часов экспериментов решил, что надо зафиксировать все полезные ссылки в одном посте для будущего меня.

Краткий алгоритм установки Tensorflow и PyTorch

Примечание: Установить Tensorflow и PyTorch можно в одном виртуальном окружении, но в статье этого алгоритма нет.

Подготовка к установке

1. Определить какая версия Python поддерживается Tensorflow и PyTorch (на момент написания статьи мне не удалось установить PyTorch в виртуальном окружении с Python 3.9.5)
2. Для выбранной версии Python найти подходящие версии Tensorflow и PyTorch
3. Определить, какие версии CUDA поддерживают выбранные ранее версии Tensorflow и PyTorch
4. Определить поддерживаемую версию cuDNN для Tensorflow – не все поддерживаемые CUDA версии cuDNN поддерживаются Tensorflow. Для PyTorch этой особенности не заметил

Установка CUDA и cuDNN

1. Скачиваем подходящую версию CUDA и устанавливаем. Можно установить со всеми значениями по умолчанию
2. Скачиваем cuDNN, подходящую для выбранной версии Tensorflow (п.1.2). Для скачивания cuDNN потребуется регистрация на сайте NVidia. “Установка” cuDNN заключается в распакове архива и заменой существующих файлов CUDA на файлы из архива

Устанавливаем Tensorflow

1. Создаём виртуальное окружение для Tensorflow c выбранной версией Python. Назовём его, например, py38tf
2. Переключаемся в окружение py38tf и устанавливаем поддерживаемую версию Tensorflow pip install tensorflow==x.x.x
3. Проверяем поддержку GPU командой

   python -c "import tensorflow as tf; print('CUDA available' if tf.config.list_physical_devices('GPU') else 'CUDA not available')"
   

Устанавливаем PyTorch

1. Создаём виртуальное окружение для PyTorch c выбранной версией Python. Назовём его, например, py38torch
2. Переключаемся в окружение py38torch и устанавливаем поддерживаемую версию PyTorch
3. Проверяем поддержку GPU командой

python -c "import torch; print('CUDA available' if torch.cuda.is_available() else 'CUDA not available')"

В моём случае заработала комбинация:

• Python 3.8.8
• Драйвер NVidia 441.22
• CUDA 10.1
• cuDNN 7.6
• Tensorflow 2.3.0
• PyTorch 1.7.1+cu101

Tensorflow и PyTorch установлены в разных виртуальных окружениях.

Итого

Польза этой статьи будет понятна не скоро: систему переустанавливаю я не часто.

Если воспользуетесь этим алгоритмом и найдёте какие-то ошибки – пишите в комментарии

CUDA (Compute Unified Device Architecture) is a parallel computing platform and programming model developed by NVIDIA. It enables developers to leverage the computational power of NVIDIA GPUs for a wide array of applications, from scientific computing to deep learning and graphics rendering. Installing CUDA on Windows 11 can seem daunting, especially if you are new to programming or working with GPUs. This comprehensive guide will walk you through the entire installation process step-by-step, ensuring a smooth and successful installation.

Prerequisites

Before you begin the installation of CUDA, it is important to ensure that your system meets the necessary prerequisites. Here’s what you need:

1. Compatible NVIDIA GPU: Check if your NVIDIA GPU supports CUDA. Most NVIDIA GPUs from the GeForce 8 series and onwards support CUDA. You can find a list of supporting GPUs on NVIDIA’s official website.

2. NVIDIA Drivers: You must have the correct NVIDIA graphics driver installed. The CUDA toolkit is compatible with certain driver versions, so ensure that your driver version meets the requirements for the version of CUDA you plan to install.

3. Windows 11: As the installation pertains specifically to Windows 11, ensure your system is updated to the latest version of Windows 11.

4. Administrator Access: You will need administrative access to install CUDA on your Windows 11 machine.

Step 1: Verify Your GPU and Current Driver Installation

The first step is to check if your NVIDIA GPU is installed correctly and has the necessary driver.

1. Open Device Manager:

   • Right-click on the Start button or press Windows + X.
   • Select Device Manager.

2. Expand Display Adapters: Look for the ‘Display Adapters’ category. Your NVIDIA GPU should be listed here.

3. Check Driver Version:

   • Right-click on your NVIDIA GPU and select ‘Properties’.
   • Navigate to the ‘Driver’ tab to find the version number.
   • You can also check for updates directly from the NVIDIA site or through GeForce Experience if it’s installed.

Step 2: Download the CUDA Toolkit

Once you’ve verified that your NVIDIA GPU and drivers are functioning correctly, the next step is to download the CUDA Toolkit.

1. Visit the NVIDIA CUDA Toolkit Download Page: Open your web browser and go to the CUDA Toolkit download page at NVIDIA’s official site.

2. Select Your Operating System: Choose ‘Windows’ from the droplist of operating systems.

3. Select Your Version: Choose the appropriate version of CUDA that is compatible with your GPU and desired software (for example, visual studio).

4. Choose Installer Type: You will have options for the installer type: ‘exe (local)’ or ‘exe (network)’. For first-time installs, it is usually best to choose the ‘exe (local)’ installer.

5. Download: Click the download button and wait for the installer to finish downloading.

Step 3: Running the CUDA Installer

After you have downloaded the CUDA Toolkit, it’s time to run the installer.

1. Locate the Installer: Navigate to your download folder and find the CUDA Toolkit installer .exe file you just downloaded.

2. Run the Installer:

   • Right-click on the installer and select ‘Run as administrator’.
   • If prompted by User Account Control, click ‘Yes’ to allow the installer to make changes.

3. Follow the Installation Wizard:

   • Choose ‘Express’ installation for a quick setup without modifying any settings.
   • Alternatively, select ‘Custom’ to customize your installation options. In this case, you can choose specific components to install or not install.

4. CUDA Toolkit Components:

   • The toolkit consists of several components including the CUDA Compiler (nvcc), CUDA Samples, libraries, and documentation.
   • It is often beneficial to install the samples as they can serve as a helpful reference.

5. Install Visual Studio Integration: If you are using Visual Studio for development, make sure to include integration options when installing.

6. Finish Installation: After selecting your components, click the Install button to begin the installation process. Wait for the installation to complete, and click Finish when prompted.

Step 4: Setting Environment Variables

After successfully installing CUDA, you need to set environment variables for it to function properly.

1. Open Environment Variables:

   • Right-click on the Start button and select ‘System’.
   • Click on ‘Advanced system settings’.
   • In the System Properties window, click on the ‘Environment Variables’ button.

2. Add CUDA Path:

   • In the ‘System variables’ section, find and select the ‘Path’ variable, and click ‘Edit’.
   • Click ‘New’ and add the following paths to the list (modify the version number as required):
     • C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.xbin
     • C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.xlibnvvp
   • Click OK to save changes.

3. Adding CUDA_HOME Variable:

   • In the Environment Variables window, under ‘System variables’, click ‘New’.
   • Set the variable name to CUDA_HOME and the value to C:Program FilesNVIDIA GPU Computing ToolkitCUDAv11.x.

4. Confirm Changes: Click OK in all dialogues to confirm your changes and close the configuration windows.

Step 5: Verifying Your Installation

It’s essential to verify that your CUDA installation was successful.

1. Open Command Prompt:

   • Press Windows + R, type in cmd, and press Enter to open the Command Prompt.

2. Check CUDA Installation:

   • Type the following command and press Enter:

     nvcc --version

   • This command should return the version of CUDA installed. If you see this, your installation has been successful.

3. Compile Sample Codes:

   • Navigate to the CUDA Samples directory, typically located at C:ProgramDataNVIDIA CorporationCUDA Samplesv11.x.
   • Open the command prompt, navigate to the sample directory, and use nvcc to compile one of the samples, such as the ‘vectorAdd’ example. Use the commands:

     cd vectorAdd
     nvcc vectorAdd.cu -o vectorAdd

   • If everything compiles without errors, you can run it by typing vectorAdd.

Step 6: Troubleshooting Installation Issues

In case you encounter any problems during installation, consider the following troubleshooting steps:

• Driver Issues: Ensure your NVIDIA drivers are updated to a version compatible with the CUDA version you’re installing. You can download the latest drivers from the NVIDIA website.

• Installation Errors: If the installer fails, check the installation log file located in the temporary folder (often found in C:UsersYourUsernameAppDataLocalTemp) for error messages.

• Environment Variables: If CUDA commands are not recognized in the Command Prompt, double-check your environment variable paths and ensure they are correctly set.

• Compatibility: Ensure that Windows 11 is compatible with the version of CUDA you are trying to install, as older versions may not fully support the newest OS.

Conclusion

Installing CUDA on a Windows 11 machine involves several steps, but by following the instructions outlined in this guide, you can ensure a successful setup. With CUDA installed, you can explore the world of parallel computing and leverage the power of your NVIDIA GPU for various applications. Remember to periodically check for updates to both your CUDA version and your NVIDIA drivers to take full advantage of new features and improvements.

As you start your journey with CUDA, consider diving into sample projects and tutorials to fully utilize the capabilities it offers. Happy coding!

Download CUDA® installer

You can download the CUDA® toolkit for your operating system from the NVIDIA® Developers Portal. We recommend using the local installer type, as it can be faster. The installer is quite large, with a size of about 3 GB. However, each LeaderGPU server has a very fast internet connection, so it doesn’t take a lot of time:

Картинка с сайта: www.leadergpu.com

Open downloads in Chrome by pressing the Ctrl + J keyboard shortcut and double-clicking on the downloaded CUDA® toolkit installer:

Картинка с сайта: www.leadergpu.com

Run CUDA® installer

Most NVIDIA® packages are self-extracted archives with installers inside. You can select a specific folder or leave the default path, then click OK:

Картинка с сайта: www.leadergpu.com

Wait a minute while the archive is being extracted:

Картинка с сайта: www.leadergpu.com

The installer will begin. In the first stage, the installer checks your hardware compatibility with the system requirements:

Картинка с сайта: www.leadergpu.com

To proceed, you must agree with EULA by clicking AGREE AND CONTINUE button:

Картинка с сайта: www.leadergpu.com

Select Express to install all available components and click NEXT:

Картинка с сайта: www.leadergpu.com

Some tools integrate with the Visual Studio IDE. If it isn’t installed on the server, the installer will issue a warning. Check the box and click NEXT:

Картинка с сайта: www.leadergpu.com

Installation may take a few minutes, which is enough time to pour yourself some coffee:

Картинка с сайта: www.leadergpu.com

After installation, you can read the summary and click NEXT:

Картинка с сайта: www.leadergpu.com

Finally, you can uncheck boxes (creating shortcut and launch utility) or leave them as they are. Exit the installer by clicking on the CLOSE button:

Картинка с сайта: www.leadergpu.com

See also:

• Install NVIDIA® drivers in Windows
• Check NVLink® in Windows
• PyTorch for Windows