Comparing the whole bill, not the GPU-hour

The hourly price is a rate. The billing unit is how your time is rounded before the rate applies. Billed by the second, ten minutes cost ten minutes. Billed by the hour, ten minutes cost an hour.

For short and uneven work, rounding can outweigh a difference in the rate. On 12 Nov 2026, the median listed H200 price here was $5.40 per GPU-hour, so an unused part of an hour is real money.

A large GPU machine takes minutes to become usable. Some providers bill those minutes and some do not. The pricing page rarely says which.

Some also set a minimum charge for each start. Both details show up only on a bill, so test with one short run before you plan around a rate.

Model files and datasets are large. If they vanish when the machine stops, you download them again on every start and pay GPU time while you wait. A volume that keeps files between machines avoids that, and it is billed by size for as long as it exists.

Ask for the price per gigabyte each month, and whether a stopped machine still bills for its disk.

Some providers charge for data leaving their network and some include it. If you move checkpoints out or serve large outputs elsewhere, the rate matters.

It is rarely on the GPU pricing page. Look under network or storage pricing and add the line to your comparison.

Write one realistic week: the hours the GPU runs, how many times it starts, the gigabytes you store and the gigabytes you move out. Price that week on each provider.

The ranking by weekly total often differs from the ranking by GPU-hour. Use the total.

Runcrate bills by the second on most GPUs and by the minute on a few. Booting is never billed. Volumes keep files between instances, and one prepaid balance pays for GPUs, models and storage.

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.