llama.cpp on the M4 Pro 64GB
llamaperf has 14 llama.cpp reports on the M4 Pro 64GB, none of them a plain single-card run with a model size and quant, so there is no median to quote. The reports are listed below with how each one ran.
14 llama.cpp reports on the M4 Pro 64GB
Newest first. A run on several cards, with speculative decoding, with part of the model in system RAM or with batched requests says so, since each describes a different setup.
- Qwen2.5-Omni · Q8_067.5 t/s
- Qwen3-VL · Q8_053.6 t/s
- Qwen3.8 · UD-IQ4_XS12.8 t/s
- Qwen3.8 · Q4_K_M11.5 t/s
- Gemma 4 · Q8_078.2 t/s
- Mistral Small 3.1 · Q4_K_M17.4 t/s
- Qwen3 · Q4_K_M77.2 t/s
- gpt-oss · MXFP473.6 t/s
- Gemma 4 · Q8_042.4 t/s
- Gemma 4 · Q4_K_M29.1 t/s
- Gemma 4 · UD-Q4_K_M54.2 t/s
- Qwen3.6 · Q4_K_M12.6 t/s
- Qwen3.6 · UD-Q4_K_M48.6 t/s
- Bonsai 2 27B · PQ2_020.5 t/s
Frequently asked
How fast is llama.cpp on the M4 Pro 64GB?
llamaperf has 14 llama.cpp reports on the M4 Pro 64GB, none of them a plain single-card run with a model size and quant, so there is no median to quote. The reports are listed below with how each one ran.