How to Install NVIDIA drivers, CUDA and CUDNN for artificial inteligence (GPU training and inference aceleration) in DLIB, TensorFlow, etc
Nvidia Graphic Card with at least 4gb of Vram for training proccess with CUDA is needed (GTX, RTX, Quadro, Tesla, etc)
Why we need Nvidia videocards and AMD not work?
Because CUDA is the industry standard for AI training and the best implementation of parallelism in GPU, also is not present on AMD videocards.
first step is download and install the latest driver for your GPU(videocard) model here:
https://www.nvidia.com/Download/index.aspx?lang=en-us
next go to nvidia page and download CUDA (cuda_10.1.243_426.00_win10):
set the buttons like the image shows and download the exe(local):
the next step is install cuda, set the checked buttons like the image shows and install in default windows dirrectory:
Optional step:
install the Nvidia videocard drivers (if not installed yet). but i recomend download and install the latest driver for your GPU(videocard), see Step 1.
finally download and install CUDNN(cudnn-10.1-windows10-x64-v7.6.5.32) from here:
https://developer.nvidia.com/rdp/cudnn-archive
unzip the cudnn-10.1-windows10-x64-v7.6.5.32.zip file and next copy all the content in the CUDA folder:
in my case C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.1
replace all files when the promt ask you:
Why not install CUDA 11.5 and CUDNN 8.3?
Simply, CUDA 10 and CUDNN 7.6 offers better perfonmance(speed) in DLIB, especially in inference time.
well, we have finished all proccess here, then go to install visual studio, compile and install DLIB for example...
Share
About the Author
You want learn a lot of AI and computer visión?
- You want learn how to make AI models for detect faces, cars, animals, parts of text, persons, machinery or anything?
- You want learn how make and run AI models in desktop and mobile?
Take practical professional artificial vision courses (desktop and mobile) in Learn Production AI: https://www.learnproductionai.tech
Comentarios
Publicar un comentario