Compute
Coming soonYour account, orchestrated.
Connect your own provider account. Upcube plans the fit, provisions instances, meters usage, and tears down cleanly. Coming soon.
- provision
- size
- meter
- teardown
Connection
Bring your own provider account.
Connect your own provider account with an API token. Upcube validates the connection, then orchestrates against it — provisioning, sizing, metering, and teardown. Compute is not yet available. As designed, Upcube would not resell GPU capacity: usage would be billed by the provider to your own account.
- 01
Connect a token
A provider API token is validated and stored for the project.
- 02
Read the status
The console shows connected or disconnected — metadata only.
- 03
Orchestrate
Provisioning, sizing, metering and teardown run against your account.
Instances
One inventory for running work.
Instances carry a name, GPU type, status, storage, and CPU shape in one table with search, filter, and sort. Statuses follow the run.
- 01
Creating
Provisioning against your connected account.
- 02
Running
Live work, metered to your account.
- 03
Restoring
Coming back after an interruption.
- 04
Deleting
Tearing down; teardown completes cleanly.
| Name | GPU type | Status | Storage | CPU |
|---|---|---|---|---|
| No instancesNo GPU instances provisioned. | ||||
Plans
Size the fit before anything runs.
Describe the workload and bound the spend. The plan fits GPU options and fetches pricing estimates at plan time — estimates from your connected account, never Upcube prices.
- 01
Describe
Model label, quantization, context tokens, batch size.
- 02
Bound
Estimated hours and a maximum hourly budget.
- 03
Fit
GPU options matched to the workload with plan-time estimates.
- 04
Approve
Provisioning stays behind an approval before anything runs.
- Model label
- 7B fine-tune
- Quantization
- FP16
- Context tokens
- 8,192
- Batch size
- 8
- Estimated hours
- 6
- Max hourly budget
- Set by you
- GPU option A80 GBComfortable
- GPU option B48 GBTight
- GPU option C24 GBInsufficient
FAQ
Does Upcube sell GPU capacity?
No. You connect your own provider account; Upcube orchestrates provisioning, sizing, metering, and teardown against it.
Where does it fit in Ethen?
Compute is separate from Gateway and Local Models. Catalog listings do not establish hosting through Compute.
Run GPU work against your own account.
Compute is coming soon. Tell us which workloads you would bring.