Freivalds Verification Overhead Measurement on n x n matrices FP32
0.48Overhead (% MatMul Cost)Freivalds verification (Kernel Contracts)
Evaluation Results
| Method | Links | |
|---|---|---|
| Freivalds verification (Kernel Contracts)n=32,768, Hardware Platform=NVIDIA H100, Software Stack=CUDA, Precision=FP32, k=10, Implementation Type=batched2026.04 | 0.48 | |
| Freivalds verification (Kernel Contracts)n=32,768, Hardware Platform=AMD MI300X, Software Stack=ROCm, Precision=FP32, k=10, Implementation Type=batched2026.04 | 0.59 | |
| Freivalds verification (Kernel Contracts)n=16,384, Hardware Platform=Apple M5, Software Stack=MPS, Precision=FP32, k=10, Implementation Type=batched2026.04 | 0.92 | |
| Freivalds verification (Kernel Contracts)n=16,384, Hardware Platform=NVIDIA H100, Software Stack=CUDA, Precision=FP32, k=10, Implementation Type=batched2026.04 | 1.06 | |
| Freivalds verification (Kernel Contracts)n=16,384, Hardware Platform=AMD MI300X, Software Stack=ROCm, Precision=FP32, k=10, Implementation Type=batched2026.04 | 1.25 | |
| Freivalds verification (Kernel Contracts)n=8,192, Hardware Platform=Apple M5, Software Stack=MPS, Precision=FP32, k=10, Implementation Type=batched2026.04 | 2.1 | |
| Freivalds verification (Kernel Contracts)n=8,192, Hardware Platform=NVIDIA H100, Software Stack=CUDA, Precision=FP32, k=10, Implementation Type=batched2026.04 | 2.5 | |
| Freivalds verification (Kernel Contracts)n=8,192, Hardware Platform=AMD MI300X, Software Stack=ROCm, Precision=FP32, k=10, Implementation Type=batched2026.04 | 4.7 | |
| Freivalds verification (Kernel Contracts)n=4,096, Hardware Platform=Apple M5, Software Stack=MPS, Precision=FP32, k=10, Implementation Type=batched2026.04 | 5.3 | |
| Freivalds verification (Kernel Contracts)n=4,096, Hardware Platform=NVIDIA H100, Software Stack=CUDA, Precision=FP32, k=10, Implementation Type=batched2026.04 | 8.1 | |
| Freivalds verification (Kernel Contracts)n=4,096, Hardware Platform=AMD MI300X, Software Stack=ROCm, Precision=FP32, k=10, Implementation Type=batched2026.04 | 14.3 | |
| Freivalds verification (Kernel Contracts)n=2,048, Hardware Platform=Apple M5, Software Stack=MPS, Precision=FP32, k=10, Implementation Type=batched2026.04 | 16.6 | |
| Freivalds verification (Kernel Contracts)n=1,024, Hardware Platform=Apple M5, Software Stack=MPS, Precision=FP32, k=10, Implementation Type=batched2026.04 | 27.8 | |
| Freivalds verification (Kernel Contracts)n=2,048, Hardware Platform=NVIDIA H100, Software Stack=CUDA, Precision=FP32, k=10, Implementation Type=batched2026.04 | 42.8 | |
| Freivalds verification (Kernel Contracts)n=2,048, Hardware Platform=AMD MI300X, Software Stack=ROCm, Precision=FP32, k=10, Implementation Type=batched2026.04 | 87.1 | |
| Freivalds verification (Kernel Contracts)n=1,024, Hardware Platform=NVIDIA H100, Software Stack=CUDA, Precision=FP32, k=10, Implementation Type=batched2026.04 | 185 | |
| Freivalds verification (Kernel Contracts)n=1,024, Hardware Platform=AMD MI300X, Software Stack=ROCm, Precision=FP32, k=10, Implementation Type=batched2026.04 | 330 |