Speedup ratio on multiple GPUs
Created by: kuke
Some important arguments:
batch_size: 32
device: GPU
first_batches_to_skip: 1
hidden_dim: 1024
learning_rate: 0.002
max_batch_num: 10
mean_var: data/global_mean_var_search26kHr
minimum_batch_size: 1
parallel: True
proj_dim: 512
stacked_num: 5Speedup on different number of GPUs:
| GPUs | Performance (frames/s) | Speedup ratio |
|---|---|---|
| 1 | 2590.32 | 1.0x |
| 2 | 3004.38 | 1.16x |
| 3 | 3183.40 | 1.23x |
| 4 | 3459.31 | 1.36x |