Set-up Windows 10 machine for Deep Learning


Set-up Windows 10 machine for Deep Learning

Install TensorFlow GPU on Windows 10 (NVIDIA graphic card)

My PC specs:
MSI GF 63 8RD
Intel Core i5 8300H CPU @2.30GHz (4 Cores, 8 Threads)
RAM 24 GB (upgraded)
Graphics: NVIDIA GeForce GTX 1050 Ti with Max-Q and Intel(R) UHD Graphics 630
Windows 10 version 21H2


Before going further, make sure that your NVIDIA Graphic card is supported CUDA and Machine learning/Deep learning capability with compute capability of higher equal 3.0 https://developer.nvidia.com/cuda-gpus


  1. Install Visual Studio Community 2015-2019

    TensorFlow in Native Windows requires Microsoft Visual C++ Redistributable for Visual Studio 2015, 2017, and 2019

    Do not install the lastest version, e.g., 2022

  2. Install NVIDIA graphics drivers

    If your graphic card is suitable for ML/DL with a suitable compute capability, you can download and install the latest version of the graphic card driver by selecting your corresponding card’s specs. Then, you can install the driver by applying default options.

    Download NVIDIA graphic driver

  3. Install CUDA resources

    Cuda Toolkit 11.8

    • Download cuDNN SDK 8.6.0 (for CUDA 11.x for Windows). You then need to extract the cuDNN files, copy and paste all files and folders into the CUDA install location. E.g., C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8

    Open the Environmental Variables and add two new paths:

    1
2

     C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\bin   
 C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\libnvvp

  4. Install Anaconda 3

    Then, open Environment variable and add three more paths (under PATH)

    1
2
3

     C:\Users\yourusername\anaconda3   
 C:\Users\yourusername\anaconda3\Library\bin   
 C:\Users\yourusername\anaconda3\Scripts
    • Create a new virtual environment with python=3.10
    1
2

     conda create --name cnn python=3.10   
 conda activate cnn

  5. Install TensorFlow 2.10

    1
2

     pip install --upgrade pip   
 pip install "tensorflow<2.11"

  6. Verify installation

    To check whether TensorFlow is installed correctly with GPU support, enter the following codes in Anaconda prompt

    1

     python -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"

    If it returns something like the following line, you have successfully installed TensorFlow with GPU support.

    [PhysicalDevice(name=’/physical_device:GPU:0’, device_type=’GPU’)]

    Or, you can even test it inside virtual environment

    1
2
3
4

     import tensorflow as tf  
 print(tf.__version__)  
 print(tf.config.list_physical_devices())  
 print("Num GPUs Available: ", len(tf.config.list_physical_devices('GPU')))

    It returns the following lines if you have your GPU in use of TensorFlow

    2.10.1

    [PhysicalDevice(name=’/physical_device:CPU:0’, device_type=’CPU’), PhysicalDevice(name=’/physical_device:GPU:0’, device_type=’GPU’)]

    Num GPUs Available: 1

  7. Fix kernel dies when running tensorflow code (additional)

    If you face an error of kernel dies when you run the TensorFlow codes The kernel appears to have died

    • Go to “C:\Program Files\Microsoft Office\root\Office16\ODBC Drivers\Salesforce\lib”
    • Locate the file, “zlibwapi.dll”
    • Copy and paste it into the CUDA toolkit folder at “C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.6\bin”

    #### Sources

    https://www.lavivienpost.com/install-tensorflow-gpu-on-windows-complete-guide/
    https://github.com/microsoft/vscode-jupyter/issues/9157
    https://www.tensorflow.org/install/pip#windows-wsl2_1