Closed ytgui closed 5 years ago
A broad ML benchmark suite for measuring performance of ML software frameworks, ML hardware accelerators, and ML cloud platforms.
https://github.com/mlperf/policies Speedup relative comparing to reference implementation
1 | 2 | 3 | 4 | 5 | 6 | 7 |
---|---|---|---|---|---|---|
Hardware | ResNet-50 | SSD-ResNet-34 | Mask-R-CNN | Transformer | NCF | Mini Go |
Pascal P100 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 |
8x Volta-V100 | 64.1 | |||||
TPU-v3.8 | 48.2 | |||||
DGX-1 | 65.6 | 30.8 | 15.5 | 62.0 | 57.2 | 93.4 |
https://mlperf.org/