cainmagi/cuda repository overview

⁠Cainmagi's Docker images

A collection of my personal customized Docker images.

⁠CUDA

The customized scripts for building Docker images with CUDA toolkit locally deployed.

• Why do we need this image?

  • The reason is that the image contains the CUDA dev files (/usr/local/cuda). Developers may need these files to build pacakges with CUDA.

• Why don't we use the official build?

  • Certainly, NVIDA has its official image like nvcr.io/nvidia/cuda:12.6.3-cudnn-devel-ubuntu24.04. However, these images have specific OS. If you are intending to work with Ubuntu, Ubi, or RockyLinux, it is fine to use the official images. However, if you want to work with other builds like Debian, there are no available official images.

• Why don't we use the debian-cuda⁠ image?

  • Well, it is indeed a Debian image. However, its version is pretty old. The script used for building that image involves some Ubuntu dependencies, which is quite unrecommended.

⁠Usage

Run the following command to pull the image:

docker pull cainmagi/cuda:debian12-cu126

After that, run the following command to launch a shell

docker run --gpus all -it --rm --shm-size=1g cainmagi/cuda:debian12-cu126 bash

⁠More details

To view the scripts used for building these images, please check

https://github.com/cainmagi/Dockerfiles/tree/cuda⁠

⁠Tags and explanations