# Elastic GPU containers

Source: https://docs.oiy.ai/docs/elastic-containers

A Docker-like workflow with resource resizing, scale-to-zero, and usage-based billing.



Oiy AI is a simple container-level elastic GPU computing service. The core workflow is: **deploy a container, adjust its resources, release idle compute, and pay for usage**.

## Docker-like workflow [#docker-like-workflow]

Configure an image, startup command, environment variables, resource allocation, and optional HTTP port. Save durable files in `/workspace`. Use a catalog template when you want a ready starting configuration.

Docker-like describes this container workflow. It does not promise a Docker daemon endpoint, Docker Compose compatibility, or arbitrary privileged host access. Image access and runtime compatibility still apply.

## Scale resources with one service action [#scale-resources-with-one-service-action]

Choose a supported GPU model, memory profile and count, together with CPU and system RAM. Resize through the console or a single API action as your workload changes. The service endpoint and persistent workspace remain associated with the service.

Changes are asynchronous and can restart processes. This is elastic resource resizing, not a promise of uninterrupted hot-resizing or a measured provisioning latency. Review [compute profiles](/docs/compute) and [service lifecycle](/docs/services).

## Scale idle compute to zero [#scale-idle-compute-to-zero]

Idle sleep releases the service's compute allocation. A request to its authenticated endpoint can wake it again. Explicit task activity protects long-running jobs from being treated as idle.

**Setting `gpuCount` to `0` is different:** it selects CPU-only compute, which can still be allocated and billed. Scale-to-zero means no compute allocation while the service sleeps. Read [scale-to-zero and wake behavior](/docs/services/sleep).

## Pay for usage [#pay-for-usage]

Compute is metered while allocated, with second-level accounting. Billing ends after the runtime confirms release. Provisioned persistent storage is billed separately and continues while compute sleeps.

Check current placement and resource prices in the console. Use [billing history](/docs/billing-history) to review posted charges, full-window summaries, and resource-level history.

## Start simply [#start-simply]

1. Sign in, verify your email, and check balance and available capacity.
2. Choose a template or runtime-approved image.
3. Set resources, deploy, and wait for the application to become ready.
4. Resize, pause, or let idle sleep release compute around your workload.

Follow the [quickstart](/docs/quickstart). Automatic horizontal replica scaling and function-based deployment APIs are outside the current preview; the implemented elastic unit is a container service.