Multi-modal to Audio Generation Latency on ServeGen 16-GPU setup (real-world workload)
21.3P50 LatencyCornserve
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CornserveInput Modality=T+I, Hardware=16-GPU2025.12 | 21.3 | 30.4 | 39.6 | |
| CornserveInput Modality=T+I+V, Hardware=16-GPU2025.12 | 21.9 | 36.7 | 44.3 | |
| CornserveInput Modality=T+I+A, Hardware=16-GPU2025.12 | 23.1 | 33.6 | 37.7 | |
| CornserveInput Modality=T+I+V+A, Hardware=16-GPU2025.12 | 24.8 | 36.5 | 37.8 | |
| MonolithInput Modality=T+I+V+A, Hardware=16-GPU2025.12 | 26.9 | 156.9 | 167 | |
| MonolithInput Modality=T+I+A, Hardware=16-GPU2025.12 | 29.6 | 182.5 | 284.5 | |
| MonolithInput Modality=T+I, Hardware=16-GPU2025.12 | 86.2 | 288.6 | 375.2 | |
| MonolithInput Modality=T+I+V, Hardware=16-GPU2025.12 | 88.8 | 236.5 | 340.5 |