A simple test: one of the most basic Keras examples slightly modified to test the time per epoch and time per step in each of the following configurations. Results below.
- Macbook Air 2020 (Apple M1)
- Dell with Intel i7-9850H and NVIDIA Quadro T2000
- Google Colab with Tesla K80
| Setup | Avg. time per epoch [s] | Avg. time per step [ms] |
|---|---|---|
| M1 GPU disabled | 21 | 33.8 |
| M1 w/ GPU | 16.6 | 26.4 |
| Colab CPU | 120.8 | 192.6 |
| Colab CPU / Tesla K80 | 10 | 16 |
| Dell i7-9850H | 24.4 | 39.2 |
| Dell i7-9850H / NVIDIA Quadro T2000 | 8.6 | 13.8 |
Check this question at Apple Developer Forums
can we expect, once multi-GPU is available for the M1, an increase in performance - maybe close to 8x if the 8 GPU cores become available, and would the GPU cores will be seen as a single GPU as the NVIDIA cards or we will need to use a distribution strategy to be able to use them in parallel?