M2 Ultra
A15 · M2 · Avalanche / Blizzard · 5 nm · First available 2023-06-13
Every configuration
| Configuration | CPU core layout | Bandwidth | GB7 CPU | GB7 Metal | Prefill Q4_0 tok/s | Decode Q4_0 tok/s |
|---|---|---|---|---|---|---|
| 24 CPU / 60 GPU (binned) | 16P + 8E | 800 GB/s | 28,386 | — | 1,014 | 88.6 |
| 24 CPU / 76 GPU (full chip) | 16P + 8E | 800 GB/s | 28,386 | 222,482 | 1,238 | 94.3 |
All benchmarks
| Benchmark | Median |
|---|---|
| Geekbench 7 · Multi-core | 28,386 (27,948–28,823, 4 sources) |
| Geekbench 7 · Metal GPU | 222,482 (222,482–222,482, 1 sources) |
| Geekbench AI · GPU · Single | — |
| Geekbench AI · GPU · Half | — |
| Geekbench AI · GPU · Quantized | — |
| Geekbench AI · Neural Engine · Single | — |
| Geekbench AI · Neural Engine · Half | — |
| Geekbench AI · Neural Engine · Quantized | — |
| Prefill 512 · F16 | 1,265 (1,129–1,402, 2 sources) |
| Decode 128 · F16 | 40.4 (39.9–41, 2 sources) |
| Prefill 512 · Q8_0 | 1,126 (1,003–1,249, 2 sources) |
| Decode 128 · Q8_0 | 64.4 (62.1–66.6, 2 sources) |
| Prefill 512 · Q4_0 | 1,126 (1,014–1,238, 2 sources) |
| Decode 128 · Q4_0 | 91.5 (88.6–94.3, 2 sources) |
Similar performance
Nearest measured configurations by Geekbench 7 CPU and Metal. Ratios above 1 are faster.
24 CPU / 76 GPU (full chip)
- M4 Max 16 CPU / 40 GPU (full chip), 2024-11: CPU 1.01×, GPU 0.92×
- M3 Ultra 28 CPU / 60 GPU (binned), 2025-03: CPU 1.23×, GPU 1.01×
- M5 Max 18 CPU / 40 GPU (full chip), 2026-03: CPU 1.23×, GPU 1.07×
- M5 Max 18 CPU / 32 GPU (binned), 2026-03: CPU 1.23×, GPU 0.87×
- M4 Max 14 CPU / 32 GPU (binned), 2024-11: CPU 0.9×, GPU 0.76×
- M3 Ultra 32 CPU / 80 GPU (full chip), 2025-03: CPU 1.38×, GPU 1.12×