| Config | TTFT mean (ms) | TTFT median (ms) | TTFT p95 (ms) | max VRAM (MiB) | max power (W) | duration (s) |
| -------- | ---------------- | ------------------ | --------------- | ---------------- | --------------- | -------------- |
| final-gemma-q4-dflash-f16 | 178.2 | 91.0 | 558.2 | 10766 | 158.0 | 29.1 |
| final-gemma-q4-dflash-q4 | 177.1 | 90.0 | 555.7 | 9730 | 164.1 | 22.8 |
| final-gemma-q4-dflash-q8 | 177.3 | 90.2 | 552.9 | 10043 | 161.4 | 24.3 |
| final-gemma-q4-mtp | 164.8 | 85.4 | 504.2 | 9135 | 164.4 | 19.9 |
| final-gemma-q4-solo | 149.0 | 83.1 | 426.6 | 8616 | 164.2 | 31.9 |
| final-gemma-q5-dflash-f16 | 187.5 | 98.2 | 573.3 | 11980 | 160.8 | 27.9 |
| final-gemma-q5-mtp | 172.7 | 92.4 | 521.2 | 10366 | 164.1 | 13.4 |
| final-gemma-q5-solo | 158.0 | 90.2 | 442.0 | 9848 | 164.3 | 18.6 |
| final-gemma-q8-dflash-f16 | 211.8 | 130.1 | 580.9 | 15888 | 149.2 | 104.6 |
| final-gemma-q8-mtp | 195.5 | 122.8 | 535.1 | 14251 | 166.6 | 77.8 |
| final-gemma-q8-solo | 178.4 | 119.8 | 445.3 | 13741 | 149.0 | 138.2 |
| final-qwen-q4-dflash | 76.7 | 36.3 | 260.2 | 9288 | 161.8 | 88.1 |
| final-qwen-q4-dspark-p0 | 76.9 | 36.6 | 259.1 | 9429 | 158.6 | 43.6 |
| final-qwen-q4-dspark-p2 | 77.0 | 36.5 | 260.4 | 9426 | 158.0 | 40.9 |
| final-qwen-q4-dspark-p4 | 77.0 | 36.6 | 260.0 | 9418 | 155.9 | 41.4 |
| final-qwen-q4-dspark-p6 | 77.3 | 36.9 | 260.8 | 9417 | 158.0 | 45.1 |
| final-qwen-q4-eagle3 | 74.8 | 35.4 | 253.8 | 8191 | 159.0 | 43.6 |
| final-qwen-q4-solo | 65.2 | 32.0 | 213.5 | 6683 | 158.2 | 53.8 |
| final-qwen-q4-vanilla17b | 79.0 | 39.3 | 253.2 | 8691 | 164.6 | 49.1 |
| final-qwen-q5-dflash | 79.5 | 38.0 | 264.3 | 9982 | 162.4 | 92.1 |
| final-qwen-q5-dspark-p0 | 79.6 | 38.0 | 266.2 | 10130 | 161.2 | 43.8 |
| final-qwen-q5-eagle3 | 78.2 | 37.1 | 258.6 | 8902 | 159.3 | 46.7 |
| final-qwen-q5-solo | 68.0 | 33.8 | 220.2 | 7396 | 164.2 | 59.3 |
| final-qwen-q8-dspark-p0 | 84.0 | 47.9 | 250.3 | 12628 | 138.4 | 57.8 |
| final-qwen-q8-eagle3 | 83.0 | 47.6 | 244.9 | 11397 | 151.3 | 53.0 |
| final-qwen-q8-solo | 70.7 | 42.1 | 203.2 | 9881 | 150.8 | 78.2 |

_All timings from clean (non-sentinel) records; VRAM/power from vram.json; duration from metrics.json._