# Match execution choices to the work you need to run

Canonical: https://savrn.com/cloud/compute

SAVRN Cloud · Coming soon. Public documentation and a browser simulation are available; connected services are not yet available.

Compute decisions involve more than selecting a GPU. A research workload also needs an appropriate runtime, a placement decision, resource limits and a way to stop spending. SAVRN Cloud’s preview demonstrates this configuration process with synthetic provider records. Connected compute and actual supplier capacity are Coming soon.


## Start from workload requirements

Identify whether the task concerns interactive inference, a repeatable evaluation or an adaptation experiment. Expected concurrency, model configuration and data handling can change the suitable execution choice. The planned service would qualify these combinations rather than infer eligibility from a hardware name. Published hardware specifications alone do not establish the performance of a particular workload.

## Review a configuration before allocation

In the preview, select a synthetic provider option, configure a model and placement, then review the allocation. The confirmation includes a simulation acknowledgement. Creating the record changes browser state and starts a sample deployment lifecycle; it does not contact a supplier, reserve a GPU or create a billable resource.

## Make lifecycle costs visible

A future compute offer needs to explain when allocation begins, which resources continue while requests are paused and what termination removes. Storage, network activity and retained artifacts can have different lifetimes from a serving process. The preview’s allocation ceiling and sample usage help illustrate these questions without publishing a supplier quote or a customer rate.

## Build toward replaceable execution

SAVRN’s plan is to support qualified rented capacity first and develop an owned-infrastructure path where it serves the workload. The project and evidence layer should survive that change. Portability still requires runtime compatibility, artifact transfer, data policy checks and measured validation at the destination. It is a design direction, not a promise that every configuration can move unchanged.

## What you can explore today

Provider selection, allocation review and deployment creation use browser-local synthetic inventory.

## Before this service launches

Connected compute requires supplier access, verified capacity, supported runtimes, resource reconciliation and documented operational responsibilities.

## The customer outcome

The intended outcome is an informed placement decision with explicit limits and lifecycle expectations before resources are committed.

## Related documentation

- [Compute overview](https://savrn.com/cloud/docs/compute-overview)
- [Rented compute preview](https://savrn.com/cloud/docs/compute-rented-digitalocean)
- [Move to owned capacity](https://savrn.com/cloud/docs/compute-owned-transition)
- [Explore compute](https://savrn.com/cloud/console/#/compute): Review a simulated configuration and allocation.