Skip to content
Cloud Provider Benchmarks

Pricing

Billing granularity compared: how per-second, per-minute and hourly billing changes short-job cost

How tracked cloud providers bill compute by the second, minute or hour, with the minimum charges and stop rules that change what short jobs cost.

Published: Updated: Türkçe oku

Direct answer

Three tracked providers that publish a per-second rule, AWS, Google Cloud and DigitalOcean, bill at least 60 seconds of compute for a run shorter than a minute. Before comparing two hourly rates, check the billing unit, the minimum, and whether billing continues after a stop.

Executive summary

Billing granularity decides what a short job costs even when the hourly rate stays the same. AWS bills Linux on-demand instances per second with a 60-second minimum. Google Cloud bills Compute Engine VMs per second after a one-minute minimum. DigitalOcean bills bundled CPU and GPU Droplets per second with a minimum of 60 seconds or $0.01, whichever is higher. PortvMind Cloud bills per minute. Vultr documents hourly GPU rates on a 730-hour month. Hetzner bills cloud servers per started hour up to a monthly cap, and Microsoft Azure's virtual machine granularity is not recorded as verified. Each rule below was checked against the provider's billing documentation on 2026-11-12.

Key findings

  1. 01 Per-second billing with a minimum makes very short runs cost the minimum, not their actual duration.
  2. 02 Stopping a resource does not always stop billing: DigitalOcean keeps billing powered-off Droplets until they are destroyed, and PortvMind Cloud keeps billing disks and stored data after a compute stop.
  3. 03 Hetzner bills Object Storage per hour with a monthly cap, a different rule from compute billing.
  4. 04 Vultr documents GPU plans on a 730-hour month, while its standard compute instances cap monthly billing at 672 hours.

Billing units by provider

Provider Billing unit Minimum or rule
AWS Per second, Linux on-demand instances 60-second minimum. SUSE Linux Enterprise Server instances are billed as a full hour.
PortvMind Cloud Per minute, pay as you go The pricing page does not say how a partial minute is rounded.
Google Cloud Per second, Compute Engine VMs One-minute minimum, then per second.
Microsoft Azure Not recorded as verified Virtual machine granularity is not confirmed in this dataset.
DigitalOcean Per second, bundled CPU and GPU Droplets 60 seconds or $0.01, whichever is higher.
Vultr Hourly GPU rates on a 730-hour month Sub-hour billing is not described in the documentation.
Hetzner Cloud servers per started hour, rounded up, capped monthly Object Storage is billed per hour with a monthly cap (see below).

What a 20-second run costs

This section is derived from the minimums above. It compares only the providers whose documentation states a minimum for compute.

  • AWS Linux on-demand instance: billed for 60 seconds.
  • Google Cloud Compute Engine VM: billed for one minute. Google’s documentation gives the same rule with a 30-second example.
  • DigitalOcean bundled CPU or GPU Droplet: billed for 60 seconds at the hourly rate, or the $0.01 minimum if that is higher.

The three providers bill a 20-second run at their minimum. The difference between them lies in the minimum and the rate, not in the run time.

No figure is given for PortvMind Cloud, Vultr or Microsoft Azure, because their documentation checked for this article does not state the rule for a run shorter than the billing unit. Hetzner rounds every started hour up, so a short cloud server run is billed for a full hour. That unit differs from the per-second providers above, so it is not placed in this comparison.

Stopping does not always stop billing

  • AWS: on-demand usage is measured from launch “until it is terminated or stopped”.
  • DigitalOcean: powered-off bundled CPU Droplets and GPU Droplets keep billing until they are destroyed.
  • Hetzner: a cloud server is billed until it is deleted, whether it is powered on or off.
  • PortvMind Cloud: stopping a resource stops the compute charge, while allocated disks and stored data keep accruing.

Object storage uses a different rule

Hetzner’s Object Storage documentation describes a base price per account. It is charged per hour with a monthly price cap, and the base price includes a storage and egress quota that accumulates each hour. Storage and traffic beyond the quota are billed separately. S3 operations such as PUT, GET and DELETE are free of charge, but the egress caused by them is billed. Objects smaller than 64 kB are charged as 64 kB.

How to use this in a comparison

  1. Note the billing unit and the minimum for the service you will use.
  2. Check whether billing continues after a stop, and what must be deleted to end it.
  3. Estimate the run pattern: how long each run lasts and how many runs happen per month.
  4. For object storage, check the minimum billable object size and the included quota.

Methodology and limitations

  • Billing rules come from each provider’s own documentation, checked on 2026-11-12. Derived values are labelled.
  • Sustained-use and committed-use discounts from Google Cloud and reservations or savings plans from other providers are outside this comparison.
  • Taxes and currencies are covered in cloud prices across currencies and tax.
  • The pricing approach is described in the pricing methodology.

Conclusion

A short run is priced by its minimum, its billing unit and its stop rule before the hourly rate matters. Check those three terms before comparing two rates.

Frequently asked questions

Is per-second billing always cheaper?

Not for very short runs. A minimum still applies, so a run shorter than the minimum is billed at the minimum. Beyond the minimum, per-second billing charges only the seconds used, which matters for runs of a few minutes.

Does stopping a server stop the bill?

It depends on the provider. DigitalOcean keeps billing powered-off Droplets until they are destroyed. PortvMind Cloud stops the compute charge when a resource is stopped, but allocated disks and stored data keep accruing.

Why is there no figure for Microsoft Azure?

The dataset does not record Azure's virtual machine billing granularity as verified, so this article does not state one.

Sources

  1. [1] Amazon EC2 On-Demand Pricing — AWS Official documentation accessed:
  2. [2] Compute Engine VM instance pricing — Google Cloud Official documentation accessed:
  3. [3] Droplet Pricing — DigitalOcean Official documentation accessed:
  4. [4] How are GPU products billed differently? — Vultr Official documentation accessed:
  5. [5] PortvMind Cloud pricing approach — PortvMind Supplied by provider team accessed:
  6. [6] Hetzner Object Storage overview — Hetzner Official documentation accessed:
  7. [7] Hetzner Cloud billing FAQ — Hetzner Official documentation accessed: