Date: 23.08.2025

NVIDIA CUDA Docker Images for PyTorch / TensorFlow

A concise snapshot of the latest minor/patch releases per major CUDA line (starting from CUDA 11), mapped to corresponding PyTorch/TensorFlow images.

• Only fixed tags (no latest).
• Images are the official pytorch/pytorch and tensorflow/tensorflow images.

CUDA	Ubuntu	Framework version	Docker image
11.2	Ubuntu 20.04	TensorFlow 2.10.0	tensorflow/tensorflow:2.10.0-gpu
11.3	Ubuntu 20.04	PyTorch 1.12.1	pytorch/pytorch:1.12.1-cuda11.3-cudnn8-devel
11.6	Ubuntu 20.04	PyTorch 1.13.1	pytorch/pytorch:1.13.1-cuda11.6-cudnn8-devel
11.7	Ubuntu 20.04	PyTorch 2.0.1	pytorch/pytorch:2.0.1-cuda11.7-cudnn8-devel
11.8	Ubuntu 22.04	PyTorch 2.7.1	pytorch/pytorch:2.7.1-cuda11.8-cudnn9-devel
11.8	Ubuntu 20.04	TensorFlow 2.14.0	tensorflow/tensorflow:2.14.0-gpu
12.1	Ubuntu 22.04	PyTorch 2.5.1	pytorch/pytorch:2.5.1-cuda12.1-cudnn9-devel
12.2	Ubuntu 22.04	TensorFlow 2.15.0	tensorflow/tensorflow:2.15.0-gpu
12.3	Ubuntu 22.04	TensorFlow 2.16.1	tensorflow/tensorflow:2.16.1-gpu
12.3	Ubuntu 22.04	TensorFlow 2.17.0	tensorflow/tensorflow:2.17.0-gpu
12.4	Ubuntu 22.04	PyTorch 2.6.0	pytorch/pytorch:2.6.0-cuda12.4-cudnn9-devel
12.5	Ubuntu 22.04	TensorFlow 2.19.0	tensorflow/tensorflow:2.19.0-gpu
12.6	Ubuntu 24.04	PyTorch 2.8.0	pytorch/pytorch:2.8.0-cuda12.6-cudnn9-devel
12.8	Ubuntu 22.04	PyTorch 2.8.0	pytorch/pytorch:2.8.0-cuda12.8-cudnn9-devel
12.9	Ubuntu 22.04	PyTorch 2.8.0	pytorch/pytorch:2.8.0-cuda12.9-cudnn9-devel

NOTES

• Avoid using the proprietary NVIDIA repository nvcr.io as most usage scenarios may violate NVIDIA’s license.