Price per GPU or per instance
Why GPU prices are compared per GPU-hour, how instance prices convert, and what a price may leave out.
Published Oct 19, 2026
GPU-hour
One GPU used for one hour. Providers sell servers (instances) with one, two, four or eight GPUs. The price per GPU-hour is the instance price per hour divided by the number of GPUs in it.
What the instance includes
Most instance prices include vCPU, memory and local storage. Some providers publish a price for the GPU only; such prices are marked «GPU only» and are not compared with full prices.
«from» prices
A price marked «from» is a minimum. The configuration you need may cost more. These prices are shown with the mark and do not count for the index.
Costs outside the GPU price
Data transfer out (egress), block and object storage, support and minimum commitments are listed separately by many providers. Provider pages list them where the provider publishes them.
Currencies
Prices are stored in the currency of the provider and converted with the reference rates of the European Central Bank. The rate date is shown with every price page.