2benchesM3
A16 / A17 · M3 · Everest / Sawtooth family · 3 nm · First available 2023-11-07
Every configuration
| Configuration | CPU core layout | Bandwidth | GB7 CPU | GB7 Metal | Prefill Q4_0 tok/s | Decode Q4_0 tok/s |
|---|---|---|---|---|---|---|
| 8 CPU / 8 GPU (binned) | 4P + 4E | 100 GB/s | 11,674 | 42,873 | — | — |
| 8 CPU / 9 GPU (binned) | 4P + 4E | 100 GB/s | — | — | — | — |
| 8 CPU / 10 GPU (full chip) | 4P + 4E | 100 GB/s | 11,610 | 47,788 | 186.8 | 21.3 |
All benchmarks
| Benchmark | Median |
|---|---|
| Geekbench 7 · Multi-core | 11,674 (11,546–11,825, 5 sources) |
| Geekbench 7 · Metal GPU | 47,646 (42,873–47,930, 3 sources) |
| Geekbench AI · GPU · Single | 8,174 (8,118–8,229, 2 sources) |
| Geekbench AI · GPU · Half | 9,386 (9,347–9,425, 2 sources) |
| Geekbench AI · GPU · Quantized | 8,643 (8,575–8,710, 2 sources) |
| Geekbench AI · Neural Engine · Single | 4,075 (4,066–4,083, 2 sources) |
| Geekbench AI · Neural Engine · Half | 30,662 (30,609–30,714, 2 sources) |
| Geekbench AI · Neural Engine · Quantized | 34,410 (34,342–34,477, 2 sources) |
| Prefill 512 · F16 | — |
| Decode 128 · F16 | — |
| Prefill 512 · Q8_0 | 187.5 (187.5–187.5, 1 sources) |
| Decode 128 · Q8_0 | 12.3 (12.3–12.3, 1 sources) |
| Prefill 512 · Q4_0 | 186.8 (186.8–186.8, 1 sources) |
| Decode 128 · Q4_0 | 21.3 (21.3–21.3, 1 sources) |
Similar performance
Nearest measured configurations by Geekbench 7 CPU and Metal. Ratios above 1 are faster.
8 CPU / 8 GPU (binned)
- M4 8 CPU / 8 GPU (binned), 2024-05: CPU 1.15×, GPU 1.1×
- M2 8 CPU / 10 GPU (full chip), 2022-06: CPU 0.83×, GPU 1.01×
- M4 10 CPU / 8 GPU (binned), 2024-05: CPU 1.29×, GPU 1.11×
- M4 10 CPU / 10 GPU (full chip), 2024-05: CPU 1.3×, GPU 1.24×
- M1 8 CPU / 8 GPU (full chip), 2020-11: CPU 0.7×, GPU 0.66×
- M3 Pro 11 CPU / 14 GPU (binned), 2023-11: CPU 1.31×, GPU 1.64×
8 CPU / 10 GPU (full chip)
- M4 8 CPU / 8 GPU (binned), 2024-05: CPU 1.16×, GPU 0.99×
- M2 8 CPU / 10 GPU (full chip), 2022-06: CPU 0.83×, GPU 0.9×
- M4 10 CPU / 8 GPU (binned), 2024-05: CPU 1.3×, GPU 0.99×
- M4 10 CPU / 10 GPU (full chip), 2024-05: CPU 1.31×, GPU 1.11×
- M2 Pro 10 CPU / 16 GPU (binned), 2023-01: CPU 1.18×, GPU 1.55×
- M3 Pro 11 CPU / 14 GPU (binned), 2023-11: CPU 1.32×, GPU 1.47×