Most instances reach a usable SSH prompt in under 60 seconds. Larger images and the H200 class can take up to two minutes. You are not billed for time spent in the queued or provisioning states.
By the second, against prepaid credit. Billing starts when the instance enters the running state and stops the moment you destroy it. There is no rounding up to the hour and no minimum charge.
No. Credit sits on your account until you spend it.
Yes, within 60 days of purchase, for credit that has not been spent. See the refund policy for the detail.
Yes. Every instance is a full machine with root, not a container. Install what you like, load custom kernel modules, run your own Docker daemon.
No. Each instance holds a dedicated physical card passed through to your machine. Nothing is time-sliced between customers.
The root volume is cryptographically erased. Anything you want to keep should live on a persistent volume or be copied out before teardown.
Yes, with persistent volumes at $0.08/GB/month. They detach and reattach to any instance in the same region.
The first 10 TB of egress each month is free, then $0.01/GB. Ingress is always free.
Multi-GPU nodes are available on the SXM classes. Use the enterprise contact form and we will configure the allocation.
You are credited for the interrupted time automatically. The SLA page sets out the thresholds.
US East, US West, EU West and AP South. Not every GPU is in every region; the catalogue page lists availability per card.
Not currently. All images are Linux. Windows licensing is available on enterprise agreements only.
Custom images are supported on enterprise accounts. On standard accounts, start from a stock image and script your setup with cloud-init.