On-demand instance
A cloud GPU you rent by the hour or second with no commitment, at the standard rate. Available immediately, billed only while running, but the most expensive per hour.
What it is
The default way to rent a GPU: start it when you need it, stop it when you do not, pay the listed per-GPU-hour price. No discount, no lock-in, no eviction risk.
Why it matters
Fine for development, spiky workloads and anything short-lived. For a service that runs 24/7, a reserved commitment (1-3 year) can cut the rate 40-60%, and spot cuts it more if the job can tolerate interruption.