Physics-Informed Neural Network Performance Comparison
| Benchmark | Points | PyTorch CPU | Rust CPU | CPU Speedup | PyTorch GPU | Rust GPU | GPU Speedup |
|---|---|---|---|---|---|---|---|
| data_generation/synthesize_displacement | 200 | 23.9 µs | - | 24.1 µs | - | ||
| data_generation/synthesize_displacement | 1000 | 34.4 µs | - | 34.4 µs | - | ||
| data_generation/synthesize_displacement | 10000 | 143.2 µs | - | 145.7 µs | - | ||
| data_generation/synthesize_displacement | 100000 | 1.25 ms | - | 1.28 ms | - | ||
| forward_pass/lffn_mlp | 200 | 113.1 µs | - | 113.7 µs | - | ||
| forward_pass/lffn_mlp | 1000 | 235.0 µs | - | 106.5 µs | - | ||
| forward_pass/lffn_mlp | 10000 | 1.60 ms | - | 106.3 µs | - | ||
| pde_residual/helmholtz | 200 | 41.2 µs | - | 42.1 µs | - | ||
| pde_residual/helmholtz | 1000 | 53.7 µs | - | 53.7 µs | - | ||
| pde_residual/helmholtz | 10000 | 176.9 µs | - | 180.5 µs | - | ||
| training_100_epochs/train_100 | 200 | 92.65 ms | - | 90.06 ms | - | ||
| training_step/single_step | 200 | 861.0 µs | - | 867.4 µs | - | ||
| training_step/single_step | 1000 | 1.17 ms | - | 876.7 µs | - | ||
| wave_number/calculate_k | 1 | 9.4 µs | - | 9.2 µs | - |