CUDA Docker environment is supported by Ubuntu nvidia cuda toolkit. Instruction: CUDA and cuDNN Install | Pop!_OS.
This docker file is used for Ubuntu 22.04 LTS, CUDA version 12.1. You may change the base system and the CUDA version listed here: nvidia/cuda | Docker Hub.
More Docker tags are in other Git branches. Tag names convention: CUDA_VERSION[-anaconda][-torch_VERSION]
.
latest
: My customized jupyter notebook11.8
: CUDA 11.8, Miniconda installed11.8-anaconda
: CUDA 11.8, Anaconda 23.07 installed11.8-torch_2.1.1
: CUDA 11.8, Anaconda with PyTorch 2.1.1
The images with Anaconda automatically run a jupyter notebook on port 8888
. Working directory in container: /work/
.
docker run --detach \
--name conda-cuda \
--restart unless-stopped \
--ipc=host \
--runtime=nvidia \
--gpus all \
-p 80:8888 \
-v your_project:/work/project \
muhac/conda-cuda:latest
You can use this notebook to check your PyTorch GPU environment.