Per-second, per-minute and per-hour billing

The hourly price is a rate. The billing unit is how the provider rounds your time before applying it. Billed by the second, a run of ten minutes and twenty seconds is charged as exactly that. Billed by the minute, it rounds up to eleven minutes. Billed by the hour, it is a full hour.

On an expensive GPU the rounding is real money. A ten-minute test on a machine billed by the hour costs six times what the same test costs billed by the second. On 12 Nov 2026, the median listed B300 price here was $9.50 per GPU-hour, so an hour you did not use is not small change.

Two details hide behind the unit. Some providers bill by the second but with a minimum charge for each start, often the first minute. And a large GPU machine can take minutes before you can use it. Whether those minutes are billed is rarely on the price page.

Ask, or test it. Start the smallest machine, wait until you can log in, stop it, and read the charge.

The unit matters most for short and uneven work: development loops, evaluation runs, and agents that start and stop machines many times a day. It matters least for a training run that lasts days, where the hourly rate is nearly the whole bill.

Inference that scales up and down sits between the two. The more often machines start and stop, the more a small unit saves.

Look for the words per second, per minute or hourly on the pricing page and in the billing documentation. If neither says, treat the price as hourly until a bill shows otherwise.

Then write the unit beside the price in your brief. Two prices in different units are not yet comparable.

Runcrate bills by the second on most GPUs and by the minute on a few. Booting is never billed. One prepaid balance pays for GPUs, models and storage, so the unit you read is the unit you are charged in.

This is one of a coordinated collection of Runcrate references. Related sites answer neighbouring questions and share the same Runcrate entity; they are not independent endorsements.