GPUDECK · WORKLOAD EVIDENCE
A paid production workload on the reference fleet, measured on GeForce RTX 4080 and GeForce RTX 5090 on the same days. One workload, its own cells — never averaged with any other workload, because cross-workload aggregation is the substitution error this project exists to expose.
| DAY | GOODPUT GeForce RTX 4080 / GeForce RTX 5090 | ATTEMPTS | COMPLETIONS | CARDS |
|---|---|---|---|---|
| 2026-08-21 | 0.37× | 308 / 1312 | 274 / 1264 | 2 / 6 |
| 2026-08-22 | 0.33× | 205 / 1331 | 174 / 1269 | 2 / 6 |
| 2026-08-23 | 0.38× | 406 / 1693 | 380 / 1647 | 2 / 6 |
| 2026-08-24 | 0.59× | 76 / 294 | 69 / 281 | 2 / 6 |
| DEPLOYMENT | $/1000 COMPLETED | PRICE $/HR | SAMPLE | PRICE PROVENANCE |
|---|---|---|---|---|
| GeForce RTX 5090 | $2.97 | $0.65 | 1295 attempts | afhubbard/gpu-prices (CC BY 4.0)… |
| RTX PRO 4500 Blackwell | $6.09 | $0.72 | 262 attempts | afhubbard/gpu-prices (CC BY 4.0)… |
The measured pair page for these deployments: side-by-side cells. Downloads: ratio cells, value board.