Calculate the effective cost of completed GPU work units for cloud GPUs. Combine billable instance hours, hourly price, supporting workload costs, completed output, and failed or interrupted work. The calculator divides total workload cost by successful output, making it easier to compare configurations whose hourly prices, throughput, or reliability differ.
The main result is the cost for one successful output unit, not merely one attempted unit. This supports fair comparisons when throughput and reliability differ.
Use the same measurement window and output definition for every option you compare.
At 1,000 billable hours, $1.20 per hour, $100 in other costs, 50,000 completed units, and 5% failed work, total cost is $1,300.00 and effective unit cost is $0.0274.
Use a repeatable output such as jobs, renders, training samples, inference requests, or processed records.
Only successful output belongs in the denominator when measuring effective unit economics.
Include them in other workload costs if they belong to the same measured workload.
Yes, if each scenario uses the same work-unit definition and measurement boundary.
Slow throughput, failures, and supporting costs can outweigh a low advertised hourly rate.
| Variable | Meaning |
|---|---|
| Billable hours | Provider-billed running time |
| Completed units | Attempted output before the failure adjustment |
| Failure rate | Share of completed work that is not useful |
| Effective unit cost | All modeled cost divided by successful output |