Jacobian Computation on KUKA iiwa 7
31SpeedupBARD (compiled)
Evaluation Results
| Method | Links | |
|---|---|---|
| BARD (compiled)Batch size=4096, Hardware=NVIDIA L40S, Compilation=torch.compile2026.05 | 31 | |
| BARD-cBatch size=4096, Hardware=NVIDIA H100 80GB HBM3, Implementation=PyTorch, torch.compile=true2026.05 | 23 | |
| BARDBatch size=4096, Hardware=NVIDIA L40S2026.05 | 12 | |
| BARDBatch size=4096, Hardware=NVIDIA H100 80GB HBM3, Implementation=PyTorch, torch.compile=false2026.05 | 8.8 | |
| Pin. C++Batch size=4096, Hardware=NVIDIA H100 80GB HBM3, Implementation=C++2026.05 | 1.7 | |
| Pinocchio C++Batch size=4096, Hardware=NVIDIA L40S, Language=C++2026.05 | 1.6 | |
| ADAMBatch size=4096, Hardware=NVIDIA L40S2026.05 | 1.1 | |
| ADAMBatch size=4096, Hardware=NVIDIA H100 80GB HBM3, Implementation=PyTorch2026.05 | 0.8 |