Templates and ports

What each of the fourteen DediGPU templates installs, which port it listens on, the first command to check it, and how to change or remove it.

A template is the software installed on top of the operating system before first boot. It runs as a systemd service under /opt/dedigpu/<template>, so you can stop it, change it or throw it away.

Template	What is installed	Listens on	First check
Bare OS + driver	NVIDIA driver 570 with the CUDA 12.8 toolkit, or ROCm 6.4 on AMD. Nothing else running.	—	`nvidia-smi`
PyTorch 2.7	CUDA 12.8 or ROCm 6.4 build, cuDNN, torchvision, torchaudio, Jupyter Lab on :8888.	Jupyter Lab :8888	`python3 -c 'import torch; print(torch.cuda.device_count())'`
TensorFlow 2.18 + JAX 0.5	CUDA 12.8, XLA, Keras 3, Jupyter Lab on :8888.	Jupyter Lab :8888	`python3 -c 'import jax; print(jax.devices())'`
Jupyter Lab	Lab on :8888 with PyTorch, JAX and TensorFlow kernels side by side.	Jupyter Lab :8888	`jupyter server list`
vLLM 0.9	OpenAI-compatible server on :8000, tensor-parallel across every card, model pulled before first boot.	OpenAI-compatible API :8000	`curl localhost:8000/v1/models`
SGLang 0.4	Structured generation with RadixAttention on :30000, model pulled before first boot.	SGLang API :30000	`curl localhost:30000/health`
Text Generation Inference 3	Hugging Face TGI on :8080 with the Messages API, model pulled before first boot.	TGI API :8080	`curl localhost:8080/info`
Ollama + Open WebUI	Ollama on :11434 and a chat UI on :3000. Name an Ollama tag such as llama3.1:70b and it is pulled first.	Ollama API :11434, Open WebUI :3000	`ollama list`
ComfyUI	ComfyUI on :8188 with ComfyUI-Manager, Flux and SDXL loaders. No checkpoints bundled — bring yours.	ComfyUI :8188	`systemctl status comfyui`
Stable Diffusion WebUI Forge	Forge on :7860 with ControlNet and the usual extensions.	Forge :7860	`systemctl status forge`
Render node	Blender 4.4 headless with Cycles and OptiX, ready for your Octane, Redshift or V-Ray licence.	—	`blender -b --version`
Docker + GPU toolkit	Docker 27 with the compose plugin and the NVIDIA Container Toolkit, or the ROCm runtime on AMD.	—	`docker run --rm --gpus all nvidia/cuda:12.8.0-base-ubuntu24.04 nvidia-smi`
Kubernetes node (k3s)	Single-node k3s with the NVIDIA GPU Operator. Join it to your cluster or run it alone.	Kubernetes API :6443	`kubectl get nodes -o wide`
Slurm worker	slurmd 24.11 with Pyxis and Enroot; give us your controller address after boot and it joins.	—	`sinfo`

Ports are public. There is no firewall in front of the server. A template that listens on a port is reachable from the Internet at first boot. Jupyter and ComfyUI start with a token or a password shown on the server page; vLLM, SGLang, TGI and Ollama start without authentication. Put them behind an SSH tunnel or a reverse proxy with a key before you share the address.

Controlling a template

systemctl status vllm          # or comfyui, jupyter, ollama, tgi, sglang, forge
journalctl -u vllm -f          # logs
/opt/dedigpu/vllm/env          # arguments (model, precision, max-model-len)
systemctl restart vllm

Serving templates and the model

vLLM, SGLang, TGI and Ollama take a model at order time and pull it before first boot; the arguments are in the template's env file. Change the model there and restart. Pre-loading a model.

Compatibility

Templates are filtered at order time by the card's vendor (ROCm builds exist for PyTorch, vLLM, TGI, Ollama and Docker on MI300X) and by the OS family (ComfyUI, Forge and the render node also exist on Windows). The templates page shows the matrix.

Controlling a template

Serving templates and the model

Compatibility

Try it on a real server.