Inference Speed Benchmarking on 720P Video 81 frames with padding 1.0 (inference)
91Inference Time (s)OSP-Next
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| OSP-NextHardware=Nvidia H200, Software Optimization=torch.compile enabled, FlashAttention3, Number of GPUs/NPUs=82026.05 | 91 | 1.42 | |
| Wan2.1 BaselineHardware=Nvidia H200, Software Optimization=torch.compile enabled, FlashAttention3, Number of GPUs/NPUs=82026.05 | 129 | — | |
| OSP-NextHardware=Nvidia H200, Software Optimization=torch.compile enabled, FlashAttention3, Number of GPUs/NPUs=42026.05 | 171 | 1.49 | |
| Wan2.1 BaselineHardware=Nvidia H200, Software Optimization=torch.compile enabled, FlashAttention3, Number of GPUs/NPUs=42026.05 | 254 | — | |
| OSP-NextHardware=Nvidia H200, Software Optimization=torch.compile enabled, FlashAttention3, Number of GPUs/NPUs=22026.05 | 335 | 1.5 | |
| OSP-NextHardware=Ascend 910C, Software Optimization=torch.compile disabled, SDPA, Number of GPUs/NPUs=42026.05 | 367 | 1.43 | |
| Wan2.1 BaselineHardware=Nvidia H200, Software Optimization=torch.compile enabled, FlashAttention3, Number of GPUs/NPUs=22026.05 | 501 | — | |
| Wan2.1 BaselineHardware=Ascend 910C, Software Optimization=torch.compile disabled, SDPA, Number of GPUs/NPUs=42026.05 | 523 | — | |
| OSP-NextHardware=Nvidia H200, Software Optimization=torch.compile enabled, FlashAttention3, Number of GPUs/NPUs=12026.05 | 620 | 1.53 | |
| OSP-NextHardware=Ascend 910C, Software Optimization=torch.compile disabled, SDPA, Number of GPUs/NPUs=22026.05 | 683 | 1.75 | |
| OSP-Next-HiF8Hardware=Ascend 950PR, Software Optimization=torch.compile disabled, SDPA, Number of GPUs/NPUs=1, Quantization=HiF82026.05 | 723 | 1.69 | |
| Wan2.1 BaselineHardware=Nvidia H200, Software Optimization=torch.compile enabled, FlashAttention3, Number of GPUs/NPUs=12026.05 | 946 | — | |
| OSP-NextHardware=Ascend 950PR, Software Optimization=torch.compile disabled, SDPA, Number of GPUs/NPUs=12026.05 | 957 | 1.27 | |
| Wan2.1 BaselineHardware=Ascend 910C, Software Optimization=torch.compile disabled, SDPA, Number of GPUs/NPUs=22026.05 | 1,197 | — | |
| Wan2.1 BaselineHardware=Ascend 950PR, Software Optimization=torch.compile disabled, SDPA, Number of GPUs/NPUs=12026.05 | 1,219 | — | |
| OSP-NextHardware=Ascend 910C, Software Optimization=torch.compile disabled, SDPA, Number of GPUs/NPUs=12026.05 | 1,362 | 1.73 | |
| Wan2.1 BaselineHardware=Ascend 910C, Software Optimization=torch.compile disabled, SDPA, Number of GPUs/NPUs=12026.05 | 2,353 | — |