M1 Ultra
A14 · M1 · Firestorm / Icestorm · 5 nm · First available 2022-03-18
Every configuration
| Configuration | CPU core layout | Bandwidth | GB7 CPU | GB7 Metal | Prefill Q4_0 tok/s | Decode Q4_0 tok/s |
|---|---|---|---|---|---|---|
| 20 CPU / 48 GPU (binned) | 16P + 4E | 800 GB/s | — | — | 772.2 | 74.9 |
| 20 CPU / 64 GPU (full chip) | 16P + 4E | 800 GB/s | — | — | 1,030 | 83.7 |
All benchmarks
| Benchmark | Median |
|---|---|
| Geekbench 7 · Multi-core | 23,418 (23,418–23,418, 1 sources) |
| Geekbench 7 · Metal GPU | 150,168 (150,168–150,168, 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 | 1,022 (875.8–1,169, 2 sources) |
| Decode 128 · F16 | 35.5 (33.9–37, 2 sources) |
| Prefill 512 · Q8_0 | 913.2 (783.5–1,043, 2 sources) |
| Decode 128 · Q8_0 | 57.8 (55.7–59.9, 2 sources) |
| Prefill 512 · Q4_0 | 901.1 (772.2–1,030, 2 sources) |
| Decode 128 · Q4_0 | 79.3 (74.9–83.7, 2 sources) |