GPU cloud cost calculator
The hourly price is only part of a GPU bill. Enter your job shape and pick a provider: the calculator applies that provider’s egress price and free tier and its billing increment (per second, per minute or per hour) to show the compute you actually pay for, the rounding overhead, the egress and the effective hourly rate. Egress prices and billing increments come from the same data behind every provider page.
Hidden costs of GPU cloud: FAQ
What are the hidden costs of GPU cloud?
Three things the hourly price hides: egress (data transferred out of the cloud, typically $0.08–$0.12 per GB at hyperscalers and free at many specialist clouds), billing granularity (a provider that bills per hour charges a 61-minute job as two hours; per-second and per-minute providers charge what you use), and idle time between jobs on machines you keep running. Storage, static IPs and support plans add on top; this calculator covers the first two.
How is GPU cloud egress charged?
Per gigabyte transferred out of the provider's network, usually after a monthly free allowance. Moving a 500 GB checkpoint set out of a $0.09/GB cloud costs $45; at a free-egress provider it costs nothing. Each provider page on GPU Finder lists its egress price and free tier, and the calculator applies them when you pick the provider.
Why does billing granularity matter?
Because short jobs get rounded up. Twenty 75-minute fine-tuning runs cost 25 hours of compute on a per-second provider but 40 hours on a per-hour provider — 60% more for the same work. Per-minute billing rounds each run to 76 minutes, which is almost exact. The rounding overhead line shows exactly what the increment costs you for your job shape.
Which price should I enter?
The on-demand node price for the whole instance (not per GPU) as listed on GPU Finder — the cheapest in-stock H100 is prefilled. For spot or interruptible capacity enter the spot price, and remember an evicted run repeats its rounding overhead when it restarts.