GPU cost per accepted output: worked examples

Compute Lens compares measured workload runs using total entered compute, setup, storage and egress costs per accepted output. It filters mismatched workload labels, currencies, quality and memory limits. All examples are fictional; no live GPU quotes are supplied.

Three fictional scenarios, calculated by the same engine as the tool. They illustrate behavior and failure cases, not customer outcomes or measured provider performance.

Compare useful outputs

Setup and rejected work change the apparent GPU bargain.

Comparable runs: 2
Excluded runs: 0
Lowest entered cost / accepted: 0.00389474 USD

Recorded workload costs
Run / provider	Currency	Total estimate	Accepted / attempts	Cost / accepted	Eligibility
run-a / Example distributed GPU	USD	0.65	80 / 100	0.008125	Meets supplied constraints
run-b / Example cloud GPU	USD	0.37	95 / 100	0.00389474	Meets supplied constraints

Run this scenario Input JSON Calculated report

Inspect every input

{
  "workload": "extraction-100",
  "model": "demo-model-fp16",
  "currency": "USD",
  "minimumPassRate": 0.8,
  "runs": [
    {
      "id": "run-a",
      "provider": "Example distributed GPU",
      "workload": "extraction-100",
      "model": "demo-model-fp16",
      "currency": "USD",
      "hourlyCost": "1",
      "setupSeconds": 600,
      "runSeconds": 1200,
      "storageCost": "0.05",
      "egressCost": "0.1",
      "accepted": 80,
      "attempts": 100,
      "memoryGB": 20,
      "availableMemoryGB": 24
    },
    {
      "id": "run-b",
      "provider": "Example cloud GPU",
      "workload": "extraction-100",
      "model": "demo-model-fp16",
      "currency": "USD",
      "hourlyCost": "2",
      "setupSeconds": 30,
      "runSeconds": 600,
      "storageCost": "0.02",
      "egressCost": "0",
      "accepted": 95,
      "attempts": 100,
      "memoryGB": 20,
      "availableMemoryGB": 24
    }
  ]
}

No accepted outputs

Zero useful output stays unestimable rather than appearing free.

Comparable runs: 1
Excluded runs: 1
Lowest entered cost / accepted: 0.00389474 USD

Recorded workload costs
Run / provider	Currency	Total estimate	Accepted / attempts	Cost / accepted	Eligibility
run-a / Example distributed GPU	USD	0.65	0 / 100	Not estimable	Below pass-rate target; No accepted output
run-b / Example cloud GPU	USD	0.37	95 / 100	0.00389474	Meets supplied constraints

Run this scenario Input JSON Calculated report

Inspect every input

{
  "workload": "extraction-100",
  "model": "demo-model-fp16",
  "currency": "USD",
  "minimumPassRate": 0.8,
  "runs": [
    {
      "id": "run-a",
      "provider": "Example distributed GPU",
      "workload": "extraction-100",
      "model": "demo-model-fp16",
      "currency": "USD",
      "hourlyCost": "1",
      "setupSeconds": 600,
      "runSeconds": 1200,
      "storageCost": "0.05",
      "egressCost": "0.1",
      "accepted": 0,
      "attempts": 100,
      "memoryGB": 20,
      "availableMemoryGB": 24
    },
    {
      "id": "run-b",
      "provider": "Example cloud GPU",
      "workload": "extraction-100",
      "model": "demo-model-fp16",
      "currency": "USD",
      "hourlyCost": "2",
      "setupSeconds": 30,
      "runSeconds": 600,
      "storageCost": "0.02",
      "egressCost": "0",
      "accepted": 95,
      "attempts": 100,
      "memoryGB": 20,
      "availableMemoryGB": 24
    }
  ]
}

The model does not fit

The cheapest entered run must still satisfy the memory requirement.

Comparable runs: 1
Excluded runs: 1
Lowest entered cost / accepted: 0.008125 USD

Recorded workload costs
Run / provider	Currency	Total estimate	Accepted / attempts	Cost / accepted	Eligibility
run-a / Example distributed GPU	USD	0.65	80 / 100	0.008125	Meets supplied constraints
run-b / Example cloud GPU	USD	0.37	95 / 100	0.00389474	Memory does not fit

Run this scenario Input JSON Calculated report

Inspect every input

{
  "workload": "extraction-100",
  "model": "demo-model-fp16",
  "currency": "USD",
  "minimumPassRate": 0.8,
  "runs": [
    {
      "id": "run-a",
      "provider": "Example distributed GPU",
      "workload": "extraction-100",
      "model": "demo-model-fp16",
      "currency": "USD",
      "hourlyCost": "1",
      "setupSeconds": 600,
      "runSeconds": 1200,
      "storageCost": "0.05",
      "egressCost": "0.1",
      "accepted": 80,
      "attempts": 100,
      "memoryGB": 20,
      "availableMemoryGB": 24
    },
    {
      "id": "run-b",
      "provider": "Example cloud GPU",
      "workload": "extraction-100",
      "model": "demo-model-fp16",
      "currency": "USD",
      "hourlyCost": "2",
      "setupSeconds": 30,
      "runSeconds": 600,
      "storageCost": "0.02",
      "egressCost": "0",
      "accepted": 95,
      "attempts": 100,
      "memoryGB": 20,
      "availableMemoryGB": 16
    }
  ]
}

A mistake worth catching

Using all attempts as the denominator makes failed outputs look productive. A run with no accepted outputs has no estimable cost per accepted output.

Use the method with your records

Total cost = hourly rate × (setup + run seconds) / 3600 + storage + egress. Divide by accepted outputs; filter mismatched workloads, currencies and memory constraints.

No live GPU inventory, automatic benchmarking, reservation, infrastructure management or verified savings.

Read the complete method, sources and input contract. Updated 2026-09-19; by the AGI Scorecard team.

Continue your review

Route LabCompare bounded supplier sequences using quality assumptions, worst-case cost and latency constraints.Proof PlanPlan a zkML handoff with measured proving time, verification cost and a pinned model/input/key manifest.