Video Generation on Latte 16 frames, 512x512
4LatencyBWCache
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BWCacheNumber of GPUs=8 × A800, Dynamic Sequence Parallelism (DSP)=Enabled2025.09 | 4 | 6.5 | |
| BWCacheNumber of GPUs=4 × A800, Dynamic Sequence Parallelism (DSP)=Enabled2025.09 | 4.5 | 5.8 | |
| TeaCacheNumber of GPUs=8 × A800, Dynamic Sequence Parallelism (DSP)=Enabled2025.09 | 5 | 5.2 | |
| TeaCacheNumber of GPUs=4 × A800, Dynamic Sequence Parallelism (DSP)=Enabled2025.09 | 5.6 | 4.66 | |
| TeaCacheNumber of GPUs=2 × A800, Dynamic Sequence Parallelism (DSP)=Enabled2025.09 | 7.3 | 3.5 | |
| BWCacheNumber of GPUs=2 × A800, Dynamic Sequence Parallelism (DSP)=Enabled2025.09 | 7.3 | 3.6 | |
| PABNumber of GPUs=8 × A800, Dynamic Sequence Parallelism (DSP)=Enabled2025.09 | 7.9 | 4 | |
| PABNumber of GPUs=4 × A800, Dynamic Sequence Parallelism (DSP)=Enabled2025.09 | 8.3 | 3.1 | |
| originalNumber of GPUs=8 × A800, Dynamic Sequence Parallelism (DSP)=Enabled2025.09 | 8.9 | 2.9 | |
| originalNumber of GPUs=4 × A800, Dynamic Sequence Parallelism (DSP)=Enabled2025.09 | 9.3 | 2.8 | |
| PABNumber of GPUs=2 × A800, Dynamic Sequence Parallelism (DSP)=Enabled2025.09 | 12.2 | 2.1 | |
| BWCacheNumber of GPUs=1 × A800, Dynamic Sequence Parallelism (DSP)=Enabled2025.09 | 14.4 | 1.8 | |
| originalNumber of GPUs=2 × A800, Dynamic Sequence Parallelism (DSP)=Enabled2025.09 | 15.1 | 1.7 | |
| TeaCacheNumber of GPUs=1 × A800, Dynamic Sequence Parallelism (DSP)=Enabled2025.09 | 17.5 | 1.5 | |
| PABNumber of GPUs=1 × A800, Dynamic Sequence Parallelism (DSP)=Enabled2025.09 | 21 | 1.2 | |
| originalNumber of GPUs=1 × A800, Dynamic Sequence Parallelism (DSP)=Enabled2025.09 | 25.9 | 1 |