You can deploy images from container registries like 🐳 Docker hub by creating a service using the create new menu in the top-right of the app.
Choose deployment, give it a name, select external image, then enter the path to the image.
Paths normally consist of [registry-url]/[account]/[image]:[tag] . Northflank defaults to Docker Hub, so you can skip the registry URL, and even the account for Docker Hub library images. You can usually just use latest for the tag, unless you need a specific version.
You will need to make sure the image is compatible with the libraries you wish to use and uses ≥ CUDA 12.0 to work on H100 GPUs.
You can find suitable images on Docker Hub, Nvidia, and other container registries.
Useful images:
nvidia/cuda:12.8.0-cudnn-runtime-ubuntu22.04nvidia/cuda:12.8.0-cudnn-devel-ubuntu22.04ollama/ollama:latestvllm/vllm-openai:latestquay.io/jupyter/pytorch-notebook:cuda12-pytorch-2.6.0quay.io/jupyter/tensorflow-notebook:cuda-tensorflow-2.17.0nvcr.io/nvidia/pytorch:25.02-py3pytorch/pytorch:2.6.0-cuda12.6-cudnn9-devel or pytorch/pytorch:2.6.0-cuda12.6-cudnn9-runtimetensorflow/tensorflow:2.18.0-gpuquay.io/jupyter/scipy-notebook:latest(CPU)quay.io/jupyter/datascience-notebook:latest (CPU)You can reference the Northflank documentation for this if you need help https://northflank.com/docs/v1/application/run/run-an-image-from-a-container-registry