Run work in your GCP project with its credits (Beta)
Beta, turned on per organization. On a connected GCP account, choose Run work here, set a monthly credit
limit, run the shown script once in Cloud Shell and choose Start running work here, or run
nodus cloud connect gcp <name> --credit-limit 500. Nodus places your single-node GPU work there first, then on
Nodus capacity as the account’s burst setting allows.
Nodus holds no key: it acts as a service account the script creates, whose roles reach only machines and disks
named for this account and only the new network’s subnets. It can list the project’s instances, so use a project
dedicated to this work if you can. Machines get no service account and no SSH key, and their network takes no
inbound connection. The read-only access is unchanged.
When Google Cloud refuses a machine for GPU quota, Nodus skips that GPU model in that region, or every region for
the all-regions quota, for an hour, and the account page links to the quota request. A zone out of machines makes
Nodus try the region’s other zones.
If a cloud ever refuses Nodus’s access while machines are still running, the account page and nodus describe
name them, so you can delete them in your cloud’s console.