GPUDECK · MEASURED PAIR · NO WINNER
These two cards ran the same paid production workloads on the same days, so their goodput can be compared as a ratio inside each workload cell. The ratio is stated as GeForce RTX 4080 over GeForce RTX 5090 and changes with the workload — which is the finding. There is no single number, no winner, and no ranking.
| WORKLOAD | DAY | GOODPUT GeForce RTX 4080 / GeForce RTX 5090 | ATTEMPTS A/B | CARDS A/B | CARD-DAYS A/B |
|---|---|---|---|---|---|
| chroma-v48-detail-svd_fp8 | 2026-08-21 | 0.39× | 42/155 | 2/6 | 2/6 |
| chroma-v48-detail-svd_fp8 | 2026-08-22 | 0.25× | 25/187 | 2/6 | 2/6 |
| chroma-v48-detail-svd_fp8 | 2026-08-23 | 0.48× | 55/157 | 2/6 | 2/6 |
| chroma-v48-detail-svd_fp8 | 2026-08-24 | 0.48× | 40/91 | 2/4 | 2/4 |
| chroma1-hd_fp8_scaled | 2026-08-21 | 0.34× | 65/104 | 2/6 | 2/6 |
| chroma1-hd_fp8_scaled | 2026-08-22 | 0.28× | 38/86 | 2/6 | 2/6 |
| chroma1-hd_fp8_scaled | 2026-08-23 | 0.53× | 34/54 | 2/6 | 2/6 |
| chroma1-hd_fp8_scaled | 2026-08-24 | 0.41× | 63/70 | 2/4 | 2/4 |
| dark_beast_krea2_fp8 | 2026-08-21 | 0.37× | 308/1312 | 2/6 | 2/6 |
| dark_beast_krea2_fp8 | 2026-08-22 | 0.33× | 205/1331 | 2/6 | 2/6 |
| dark_beast_krea2_fp8 | 2026-08-23 | 0.38× | 406/1693 | 2/6 | 2/6 |
| dark_beast_krea2_fp8 | 2026-08-24 | 0.59× | 76/294 | 2/6 | 2/6 |
| dark_beast_krea2_identity_edit_v1_2 | 2026-08-21 | 0.46× | 120/390 | 2/6 | 2/6 |
| dark_beast_krea2_identity_edit_v1_2 | 2026-08-22 | 0.44× | 177/518 | 2/6 | 2/6 |
| dark_beast_krea2_identity_edit_v1_2 | 2026-08-23 | 0.43× | 131/417 | 2/6 | 2/6 |
| dark_beast_krea2_identity_edit_v1_2 | 2026-08-24 | 0.44× | 144/58 | 2/6 | 2/6 |
| dark_beast_z_image_turbo_v9_bf16 | 2026-08-21 | 0.44× | 80/290 | 2/6 | 2/6 |
| dark_beast_z_image_turbo_v9_bf16 | 2026-08-22 | 0.34× | 113/424 | 2/6 | 2/6 |
| dark_beast_z_image_turbo_v9_bf16 | 2026-08-23 | 0.33× | 108/295 | 2/6 | 2/6 |
| dark_beast_z_image_turbo_v9_bf16 | 2026-08-24 | 0.30× | 95/252 | 2/4 | 2/4 |
| krea2_identity_edit_sogni_v0_3_alpha | 2026-08-21 | 0.33× | 24/57 | 2/6 | 2/6 |
| krea2_identity_edit_sogni_v0_3_alpha | 2026-08-22 | 0.37× | 32/85 | 2/6 | 2/6 |
| krea2_identity_edit_sogni_v0_3_alpha | 2026-08-23 | 0.37× | 86/153 | 2/6 | 2/6 |
| krea2_identity_edit_sogni_v0_3_alpha | 2026-08-24 | 0.51× | 18/34 | 2/5 | 2/5 |
| krea2_identity_edit_v1_2 | 2026-08-21 | 0.33× | 256/762 | 2/6 | 2/6 |
| krea2_identity_edit_v1_2 | 2026-08-22 | 0.34× | 244/813 | 2/6 | 2/6 |
| krea2_identity_edit_v1_2 | 2026-08-23 | 0.42× | 165/718 | 2/6 | 2/6 |
| krea2_identity_edit_v1_2 | 2026-08-24 | 0.56× | 41/123 | 2/6 | 2/6 |
| krea2_turbo_fp8_scaled | 2026-08-21 | 0.40× | 455/937 | 2/6 | 2/6 |
| krea2_turbo_fp8_scaled | 2026-08-22 | 0.30× | 474/1364 | 2/6 | 2/6 |
| krea2_turbo_fp8_scaled | 2026-08-23 | 0.42× | 628/1131 | 2/6 | 2/6 |
| krea2_turbo_fp8_scaled | 2026-08-24 | 0.58× | 60/192 | 2/6 | 2/6 |
| one_obsession_v22_fp16 | 2026-08-21 | 0.21× | 229/181 | 2/6 | 2/6 |
| one_obsession_v22_fp16 | 2026-08-22 | 0.20× | 244/453 | 2/6 | 2/6 |
| one_obsession_v22_fp16 | 2026-08-23 | 0.34× | 88/192 | 2/6 | 2/6 |
| one_obsession_v22_fp16 | 2026-08-24 | 0.35× | 101/124 | 2/4 | 2/4 |
| z_image_bf16 | 2026-08-21 | 0.51× | 12/15 | 2/4 | 2/4 |
| z_image_bf16 | 2026-08-22 | 0.51× | 12/15 | 2/4 | 2/4 |
| z_image_bf16 | 2026-08-23 | 0.45× | 39/33 | 2/6 | 2/6 |
| z_image_bf16 | 2026-08-24 | 0.51× | 40/22 | 2/6 | 2/6 |
| z_image_turbo_bf16 | 2026-08-23 | 0.31× | 25/72 | 2/6 | 2/6 |
| z_image_turbo_bf16 | 2026-08-24 | 0.28× | 32/57 | 2/4 | 2/4 |
Bring your own hourly quotes to /quote to turn any of these ratios into a break-even price for your workload. Raw cells: gpu-goodput-ratios.csv.