Matrix Multiplication on Simulated Matrix Multiplication Sequence Lengths 1K-32K
27.63Latency (ms)NeuronMLP
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| NeuronMLPOptimization=Block-aligned SVD + TrainiumFusion2025.10 | 27.63 | 2.47 | 99.21 | 85.2 | 2.18 | 22.55 | |
| NKI XUVDescription=Matmuls using low-rank factors U and V from SVD, Implementation=AWS Official [6]2025.10 | 37.47 | 6.52 | 81.28 | 65.24 | 2.18 | 78.92 | |
| NKI XWDescription=Standard matmul without SVD, Implementation=AWS Official [6]2025.10 | 57.89 | 9.93 | 78.52 | 64.09 | 2.96 | 68.01 |