GPU economics

GPU Burst Billing Calculator

The cost of a burst GPU deployment depends on billed session time, not only useful inference time. Model loading, provider-billed warm idle, minimum charges and billing increments can make a short job consume a longer paid interval. This calculator applies one minimum and rounding rule per session, adds persistent costs separately, and compares that total with one always-on worker over the same window. The two workers must execute the same useful work in the same active time for the comparison to be meaningful.

Open the live GPU Burst Billing calculator - Billable versus useful GPU worker-hours →

Example scenario

Original illustrative scenario: 1,000 sessions each do 20 useful seconds, with 30 billed startup seconds and 10 billed idle seconds. A 60-second session minimum and one-second increments produce 16.667 billable worker-hours but only 5.556 useful worker-hours. At a hypothetical $3 per hour plus $20 persistent costs, bursts cost $70. An equivalent $2-per-hour worker kept for 730 hours plus $20 extras costs $1,480. Neither price nor startup duration is a provider quote.

What the inputs mean

Billing sessions in the window: Enter sessions. The initial value is illustrative, not an observed price or performance measurement.
Useful work per session: Enter seconds. The initial value is illustrative, not an observed price or performance measurement.
Billable startup per session: Enter seconds. The initial value is illustrative, not an observed price or performance measurement.
Billable warm idle per session: Enter seconds. The initial value is illustrative, not an observed price or performance measurement.
Minimum billable session: Enter seconds. The initial value is illustrative, not an observed price or performance measurement.
Billing increment: Enter seconds. The initial value is illustrative, not an observed price or performance measurement.
Complete burst worker hourly rate: Enter USD / worker-hour. The initial value is illustrative, not an observed price or performance measurement.
Burst storage, plan and other costs: Enter USD / window. The initial value is illustrative, not an observed price or performance measurement.
Comparison window: Enter hours. The initial value is illustrative, not an observed price or performance measurement.
Complete always-on worker hourly rate: Enter USD / worker-hour. The initial value is illustrative, not an observed price or performance measurement.
Always-on storage and other costs: Enter USD / window. The initial value is illustrative, not an observed price or performance measurement.

What the result means

The output exposes paid time that produced no useful work and keeps persistent charges visible. It rejects sessions that cannot fit sequentially in the selected window. Concurrent sessions, autoscaler behavior, interruptions, promotional credits and tiered tariffs require a different or explicitly split scenario.

Assumptions

A session is a provider-defined billing interval, not necessarily one HTTP request. Enter only startup and idle time that is billed. All sessions run sequentially on one complete worker. The two options must deliver the same work in the same active time. Minimum billing and rounding rules vary; use this calculator only for a minimum followed by whole billing quanta. Extras include persistent storage and other charges even at zero use.
All inputs must describe the same comparison window. Unknown costs are not automatically zero.
Default numbers are original scenarios; this calculator makes no network requests.

Where the prices come from

Use the GPU explorer to find source-attributed observations, then check purchase unit, region and terms. Rates here are editable assumptions and are never silently replaced.

Formula and methodology

Per-session billable seconds = ceil(max(minimum seconds, useful seconds + billed startup seconds + billed idle seconds) / billing increment) x billing increment. Burst cost = sessions x per-session billable seconds / 3,600 x complete-worker hourly price + burst extras. Always-on cost = comparison hours x complete-worker hourly price + always-on extras. Useful time excludes startup and idle. Unit cost is undefined without useful work; persistent extras still apply at zero sessions.

Interpretation guide

Compare alternatives with the same workload assumptions.
Stress-test output-heavy, retry-heavy, cache-miss, and power-user cases before committing budget.
Verify source links and production logs before using the estimate for billing decisions.

Limitations before production billing decisions

Treat ByteCosts calculations as planning estimates, not final billing totals. Real invoices can differ because token mix, retry rate, cache hit rate, rate limits, taxes, gateway fees, regional pricing, and negotiated discounts change the effective cost.

Verify the provider source before production billing decisions, then compare the estimate with your own logs or invoice once production traffic is live.

GPU Burst Billing Calculator. ByteCosts. https://bytecosts.com/tools/gpu-burst-billing/