Per-Second GPU Providers

Which GPU providers bill by the second?

Some GPU providers state on their own pricing page that they bill by the second. Others list an hourly rate and do not say how finely they count. The table checks each provider against its own page.

6 of the 12 GPU providers on rent.dev's price board state per-second billing on their own pages. The lowest rate among them is $0.12 per GPU hour, for the NVIDIA RTX A2000 (RunPod, community), read .

Compare GPU rates by vendor

rent.dev bills its own sandboxes by the second. Its GPU pods run on your own provider key, and that provider bills you.

Each GPU provider's billing unit, from its own page

One row per provider and source page, providers whose page states per-second billing first. Billing sentences are quoted from the source page on the day it was checked; rates per GPU hour are from rent.dev's price board, generated .
ProviderBilling unitWhat its page saysLowest per GPU hourForSourcePage checkedRate read
RunPodper second"billed per second with no commitment"$0.12NVIDIA RTX A2000runpod.io
Beamper second"billed by the second"$0.44RTX 4090 machinebeam.cloud
Daytonaper second"All billing is calculated per second."$0.57Nvidia RTX 4090daytona.io
Modalper second"Functions and Sandboxes, billed by the second for CPU, memory, and GPU time."$0.5904Nvidia T4modal.com
Northflankper second"On demand, billed by the second."$0.80NVIDIA L4northflank.com
DigitalOceanper second"GPU Droplets are billed per second with a 5-minute minimum billing period."$1.91AMD Instinct MI300Xdigitalocean.com
Akashbilling unit not on the source page$0.02gtx1050 2Gi PCIeconsole-api.akash.network
Verdabilling unit not on the source page$0.122Tesla V100 16GBapi.datacrunch.io
AWSbilling unit not on the source page$0.37853g4adpricing.us-east-1.amazonaws.com
CoreWeavebilling unit not on the source page$0.78375NVIDIA L40coreweave.com
Oraclebilling unit not on the source page$1.275X7apexapps.oracle.com
Azurebilling unit not on the source page$3.399625A100 SXM4 40GBprices.azure.com

Why the billing unit matters

Per-second billing matters most for short GPU jobs. A test that runs for a few minutes costs a few minutes, not a full hour. The words to look for are on the provider's own page: a sentence that says it bills by the second, and any minimum charge per start. An hourly price alone does not tell you the billing unit, because most providers that meter by the second still quote their rates per hour.

How the table was checked

For each provider on rent.dev's price board, we opened the page its rates come from and looked for a sentence about the billing unit. Where we found one, the table quotes it word for word and shows the day we checked.

Where the page has no such sentence, the row says the billing unit is not on the source page. That does not mean the provider bills by the hour. It means its own page does not say.

Some rates come from a data feed rather than a web page. A feed lists prices but rarely says how the meter counts, so those rows usually carry no billing sentence. Where the feed cannot be opened without a key, we read the provider's own pricing page instead, and the row links that page as its source. A row marked as not checked has a source we have not opened yet.

Minimums matter as much as the unit

A provider can meter by the second and still charge a minimum each time a machine starts. Read the whole quote: a minimum of a few minutes turns many short jobs into many minimum charges.

Read on: How per-second GPU billing works, and when it saves money.

Run a per-second sandbox

Agents can start at rent.dev/llms.txt, which lists every route.