Fine-tune on a slice of a GPU. Pay for the minutes you use.
QuickPods are one-click AI stacks on fractional NVIDIA GPUs, billed per GPU-minute, for fine-tuning, notebooks and experiments.
QuickPods are one-click AI stacks on fractional NVIDIA GPUs, billed per GPU-minute. Pick a template, choose how much of a GPU you need, and you are in a running environment before your coffee is ready.
QuickPods at a glance
What QuickPods does
Fractional GPUs
Allocate 12.5 to 100 percent of an NVIDIA GPU through PodVirt. One physical GPU serves several users, which is why a small job does not have to rent a whole card.
One-click stacks
Templates for fine-tuning, vLLM serving, ComfyUI and Jupyter come with ports, environment and disk already wired. Launch a working stack, not a bare machine.
Persistent storage and networking
Pods keep their volumes between runs and expose the ports you declare, so a notebook or a demo endpoint survives a restart.
CLI, SDK and API
Everything the console does is available from the podstack CLI, the Python SDK and the REST API, so pods fit into scripts and CI.
Built-in MLOps
Experiments, runs and model versions are tracked alongside the pod, so a fine-tune leaves a record you can compare and promote.
Live on Podstack Cloud
21 one-click templates ready · 2 GPUs available now
LLM & Fine-tuning
Large language model training and fine-tuning
LLM Inference
Large language model serving and inference
Image Generation
AI image generation and editing
ML Frameworks
Machine learning frameworks and tools
Scientific Computing
Scientific simulation and research
3D & Rendering
3D rendering and visual computing
Development Tools
Development environments and utilities
From the terminal
One CLI, one SDK, one API
$ podstack gpu launch
? pick GPU › H100 80GB · 50% slice
✓ ssh ready · billed per GPU-minute
$ podstack train create --budget 25
→ LoRA job started · adapter saved when it finishes
$ podstack models list
llama-3-70b · mistral-7b · sdxl · +18How QuickPods works
- 1
Pick a template
Choose fine-tuning, inference, ComfyUI or a plain notebook. Each carries sane defaults you can override.
- 2
Choose GPU type and share
Select the GPU family and how much of it you need, from an eighth to the whole card.
- 3
Launch and connect
The pod starts in seconds. Connect through the web terminal, Jupyter or SSH.
- 4
Stop when done
Stopping the pod stops the meter. Billing is per GPU-minute, so an afternoon of experiments costs an afternoon.
How QuickPods bills
per GPU-minute
Metered by the minute against your Podstack wallet. Rates depend on GPU type and share; the console shows the estimate before you launch.
See pricingQuestions about QuickPods
What is a QuickPod?
A QuickPod is a running AI environment on a fractional NVIDIA GPU, launched from a one-click template and billed per GPU-minute.
How small a GPU share can I run?
You can allocate as little as 12.5 percent of a GPU, which is enough for many fine-tuning runs, notebooks and small inference tests.
Which templates are available?
Templates cover fine-tuning, vLLM inference serving, ComfyUI image workflows and Jupyter notebooks, with more added as they are requested.
How am I billed for a QuickPod?
Billing is per GPU-minute from your Podstack wallet, and the meter stops the moment you stop the pod.
Can I use the CLI instead of the console?
Yes, the podstack CLI, the Python SDK and the REST API expose everything the console does, including launching and stopping pods.
What makes fractional GPUs possible?
PodVirt, the virtualisation layer in Podstack OS, splits one physical GPU into isolated shares that several users can run on at once.
Is my data isolated from other users on the same GPU?
Each pod runs in its own isolated environment with its own storage, and the platform is ISO 27001 certified.
Start with QuickPods
QuickPods are one-click AI stacks on fractional NVIDIA GPUs, billed per GPU-minute, for fine-tuning, notebooks and experiments.