Compare reserved GPU contracts before committing

A reservation is worth comparing when you know the configuration and usage you need. Compare the total commitment with the on-demand bill for that same usage, then check what happens if your plans change.

Match the configuration first

Use the same GPU variant, node size and required region. A reserved rate for a larger node cannot be compared with a single-GPU rate without including the whole node. The offer tables show per-GPU rates and minimum node totals separately.

Record whether the commitment is a calendar term, a minimum number of GPU-hours, or a quote whose minimum is not published. A low starting price without those terms is not enough to calculate a commitment total.

Calculate the cost of unused capacity

For a plan with a known total price, divide that total by the matching on-demand node rate. This gives the on-demand hours costing the same amount. Compare that number with the hours you expect to use, including the time your job loads data and writes checkpoints.

For a volume commitment, calculate GPU-hours as GPU count multiplied by runtime. Confirm which hardware can consume the committed volume and when unused volume expires. Do not infer transferability between GPU models from a discount percentage.

Check the contract details the price table cannot settle

Before booking, obtain the payment schedule, cancellation terms, capacity guarantee, eligible regions and rules for replacing failed hardware. A price observation does not establish any of those rights. Ask how storage and outgoing data are billed during and after the reservation.

Compare published commitment discounts, then use the job cost calculator to include the applicable charges. If a term is unpublished, leave it unknown until the provider confirms it.