Step By Step Installation Of Cuda And CuDNN On Windows
HTML-код
- Опубликовано: 24 июн 2020
- This video is an installation guide to Nvidia CUDA Development Kit version 10.0.130 and Nvidia CUDNN version 7.6.4 on Windows 10 machines.Since CUDA does not have it's own C++ compiler we use Visual Studio 2017 to compile nvidia programs written in C++.
CUDA(Compute Unified Device Architecture) is a parallel computing platform and an API model created by Nvidia which allows data scientists and machine learning engineers to use a CUDA-enabled accelerated graphics processing. The processing depends on the available memory and the no of cuda cores present in the GPU.
The Nvidia CUDA Deep Neural Network Library is a GPU accelerated library for deep neural networks which provides highly tuned implementation for standard routines such as forward and backward propagation, convolution, pooling, normalization and activation layers.
The following installation guide is tested on:
Python Version : 3.7.3
Operating System: Windows 10
Nvidia Cuda Driver Version: 446.14(Nvidia Cuda 11.0.140 Driver)
You can also test with Nvidia Cuda Driver version greater than or equal to the toolkit version you are installing
Visual Studio 2017
Visual Studio Code to train mnist dataset on gpu
Tensorflow with GPU support version 2.0.0
Install tensorflow 2.0.0 with gpu support using:
pip install tensorflow-gpu==2.0.0
CUDA Compatibility With Nvidia Drivers:
docs.nvidia.com/pdf/CUDA_Comp...
CUDA And CUDNN Compatibility With Tensorflow Versions
www.tensorflow.org/install/so...
Steps Involved During the Installation
1) Install Visual Studio 2017 from www.techspot.com/downloads/62...
2)
Install Cuda Development Kit 10.0.130_411.31_win10.exe from developer.nvidia.com/cuda-10....
3) Install Cudnn 7.6.4 from
developer.nvidia.com/rdp/cudn...
4) Edit environment variables and add cuda to path
5) Open cmd and install tensorflow-gpu==2.0.0
6) Test and check for cuda installation using:
from tensorflow.python.client import device_lib
print(device_lib.list_local_devices())
Finally train the neural network on mnist dataset.
Github repository: github.com/prabhat-123/Mnist_...
Goto the link and copy all the source code of "mnist.py" file and use it in your file and train the mnist dataset on your gpu..
If you face any errors or encounter any problems feel free to ask..I am always there to help you out..
Thank you Наука
An EXCELLENT video detailing the complex steps to get GPUs running with CUDA and CuDNN (for ML). I recommend anyone to follow this slowly and in detail to get your GPU / CUDA system running again (a Microsoft Windows10 forced update completely destroyed my original OS and install, this video has saved me!). Must detail your own system, GPU, python version, VisualStudio version, CUDA version, CuDNN version, tensorflow-gpu version before setting out on the lengthy install process. Thank you so very much...5stars...
Thank you for this detailed and clear tutorial brother. I was really struggling with cuda issues on my desktop and your video helped a lot in solving those problems. 🥂🤝
Thanks man, I used your tutorial to install cuda 10.5 Works perfect.
Great video and wonderful explanation. Make more videos brother. A big thumbs up
Great video. This worked for me. Thanks!
Thank you so much that i was looking for the cudnn link all were to sign up but this worked great thank you so much
Thank you for your feedback...Subscibe my channel for more AI videos...
Great job raja
Thanks bro :)
When i go on this link for Visual studio, i get 2019 instead 2017 version. What is problem, can I use 2019 version? Thanks a lot!
Yes it works for 19 as well...you have to change something's later on in the video...meaning that where I have used 17 as a reference you should replace by 19
@@prabhatale1135 Than you :)