Per-Second Billing for RunPod GPUs | Generated by AI

Home 2026.06

Question: On RunPod, for a single B200 GPU, is billing based on minutes or hours? What’s the minimum unit used to calculate cost?

Answer:

RunPod bills per-second, not per-minute or per-hour. The advertised rate is quoted per-hour, but the actual charge accrues by the second for the time your pod is running.

A few specifics that matter for a B200 single-GPU pod:

So the minimum unit is 1 second. Practical implication for your workflow: spin up, run the job, and tear down via the API/CLI rather than leaving it parked — a B200 idling for an hour between REPL iterations is real money. Script the lifecycle (runpodctl or the GraphQL API) so the pod only exists during active work.

Verify the current B200 hourly rate on the pricing page since GPU rates shift, but the billing granularity (per-second) is stable RunPod policy.

References:


Back Donate