Apple M2 Ultra (76-core GPU) × Llama 2 7B
Llama 2 7B on Apple M2 Ultra (76-core GPU)
109best gen tok/s
1,561best prompt tok/s
640max context tested
12results
Measured · all numbers below come from the linked sources
llama.cpp
| Quant | Gen tok/s | Prompt tok/s | Source | Context tested | TTFT | Build |
|---|---|---|---|---|---|---|
| Q4_0 | 109 | 1,412 | llama.cpp (Metal) ↗ | 512+128 | — | llama.cpp Metal |
| Q4_0 | 109 | 1,392 | llama.cpp (Metal) ↗ | 512+128 | — | llama.cpp Metal |
| Q4_0 | 94.3 | 1,238 | llama.cpp (Metal) ↗ | 512+128 | — | llama.cpp Metal |
| Q4_0 | 94.3 | 1,238 | llama.cpp (Metal) ↗ | 512+128 | — | llama.cpp Metal |
| Q8_0 | 73.3 | 1,387 | llama.cpp (Metal) ↗ | 512+128 | — | llama.cpp Metal |
| Q8_0 | 73.1 | 1,368 | llama.cpp (Metal) ↗ | 512+128 | — | llama.cpp Metal |
| Q8_0 | 66.6 | 1,249 | llama.cpp (Metal) ↗ | 512+128 | — | llama.cpp Metal |
| Q8_0 | 66.6 | 1,249 | llama.cpp (Metal) ↗ | 512+128 | — | llama.cpp Metal |
| F16 | 43.2 | 1,561 | llama.cpp (Metal) ↗ | 512+128 | — | llama.cpp Metal |
| F16 | 43.1 | 1,526 | llama.cpp (Metal) ↗ | 512+128 | — | llama.cpp Metal |
| F16 | 41.0 | 1,402 | llama.cpp (Metal) ↗ | 512+128 | — | llama.cpp Metal |
| F16 | 41.0 | 1,402 | llama.cpp (Metal) ↗ | 512+128 | — | llama.cpp Metal |
Ollama
No measured Ollama results yet. Submit yours →
MLX
No measured MLX results yet. Submit yours →
"Context tested" is the largest prompt+generation length actually benchmarked (tokens), not a memory-fit estimate.