> See the full index of documentation pages at https://docs.zettabyte.com/llms.txt.

# How zCLOUD works

> How instances, tasks, images and storage fit together in the zCLOUD console.

Source: https://docs.zettabyte.com/docs/zcloud/introduction/how-zcloud-works

Last updated: 2026-11-05

zCLOUD, the GPU cloud, runs work on GPU capacity you launch as instances. Tasks draw on those GPUs, start from an image, and can use storage for data.

## The two workspaces

The console at console.zettabyte.com has two workspaces, chosen in **Switch Platform**:

- zCLOUD, the **AI Infrastructure Rental** workspace, where you provision compute and run work: Tasks (Training, Fine-Tuning, Inference, IDE) and Resources (Instance, Storage Management, Models, Datasets, Images).
- zSTUDIO, listed in the platform switcher as **MaaS via API**, which has its own documentation.

## Instances vs. tasks

An Instance is a block of GPU capacity. Open [**Instance**](https://console.zettabyte.com/instance/list) under Resources to list instances; **Launch Instance** adds one, and the row menu offers **Terminate**.

A Task is a unit of work created from the Tasks menu. A task that asks for GPUs takes them from the idle GPUs on the account's instances, so launch an instance first unless the account already has idle GPUs. Tasks come in four kinds:

- Training, a job with a start and an end (run a script, train a model, exit). See [Run a training job](https://docs.zettabyte.com/docs/zcloud/tasks/run-a-training-job).
- Fine-Tuning, for adapting an existing model. See [Fine-tune a model](https://docs.zettabyte.com/docs/zcloud/tasks/fine-tune-a-model).
- Inference, a long-running service that exposes your model over an API endpoint. See [Deploy a model](https://docs.zettabyte.com/docs/zcloud/deploy-and-serve/deploy-a-model).
- IDE, an interactive development environment; the **Type** field offers Jupyter and VSCode. See [Notebooks and IDEs](https://docs.zettabyte.com/docs/zcloud/compute/notebooks-and-ides).

## Models

**Models** has two tabs, **Pre-Trained Model** and **Custom Model**. **Deploy** on a pre-trained model card opens the Create Inference Task wizard, and **Fine Tuning** opens the Create Fine-Tuning Task wizard. [Deploy a model](https://docs.zettabyte.com/docs/zcloud/deploy-and-serve/deploy-a-model) covers the wizard.

## Images

An image is the container that carries code and environment onto a GPU. The Training, IDE, and Inference task forms offer a **Preset Image** or a **Custom Image**. See [Notebooks and IDEs](https://docs.zettabyte.com/docs/zcloud/compute/notebooks-and-ides) and [Container images](https://docs.zettabyte.com/docs/zcloud/compute/container-images) for each.

## Storage

The Storage block of a task form offers three kinds of storage:

- Storage Volume (`PVC`, a persistent volume). Create one with **Add PVC** on the **PVC Storage Management** tab (open [**PVC Storage Management**](https://console.zettabyte.com/storage-manage/index)), then add it to a task with **+ Storage Volume**. Each mount path must be unique.
- Ephemeral Volume, added with **+ Ephemeral Volume**; the form takes a mount directory and a size.
- `S3`, added with **+ S3**. The **S3 Object Storage** tab of Storage Management lists buckets.

See [Storage and PVCs](https://docs.zettabyte.com/docs/zcloud/models-data-storage/storage) for the full model.

## Priority and the resource queue

Task forms carry a **Priority** setting with options 6 to 1; its hint reads "Higher values indicate higher priority". See [The resource queue](https://docs.zettabyte.com/docs/zcloud/compute/resource-queue).

## The order to create things

1. An instance, unless the account already has idle GPUs. See [GPU instances](https://docs.zettabyte.com/docs/zcloud/compute/gpu-instances).
2. A `PVC` with **Add PVC**, if the task needs a Storage Volume. See [Storage and PVCs](https://docs.zettabyte.com/docs/zcloud/models-data-storage/storage).
3. Choose an image in the task form: a preset, or a custom one. See [Container images](https://docs.zettabyte.com/docs/zcloud/compute/container-images).
4. Create the task: [Run a training job](https://docs.zettabyte.com/docs/zcloud/tasks/run-a-training-job), [fine-tune a model](https://docs.zettabyte.com/docs/zcloud/tasks/fine-tune-a-model), deploy an inference endpoint (step 5), or open an IDE.
5. To serve a model, deploy it as an inference endpoint and call it over the API. See [Deploy a model](https://docs.zettabyte.com/docs/zcloud/deploy-and-serve/deploy-a-model) and [Call your endpoint](https://docs.zettabyte.com/docs/zcloud/deploy-and-serve/call-your-endpoint).
6. **Terminate** instances that are no longer needed. See [Stop work and release a GPU instance](https://docs.zettabyte.com/docs/zcloud/compute/stop-work-and-release).

## Next steps

- [Quickstart: launch a GPU and deploy a model](https://docs.zettabyte.com/docs/zcloud/introduction/quickstart): follow the steps from sign-in to a deployed endpoint.
- [GPU instances](https://docs.zettabyte.com/docs/zcloud/compute/gpu-instances): launch and manage the capacity the quickstart uses.
- [Stop work and release a GPU instance](https://docs.zettabyte.com/docs/zcloud/compute/stop-work-and-release): end an instance when its work is done (**Terminate** in the console).
- [Storage and PVCs](https://docs.zettabyte.com/docs/zcloud/models-data-storage/storage): the full storage model.
