llama.cpp on the M5 Max 48GB
llamaperf has 38 llama.cpp reports on the M5 Max 48GB, 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.
38 llama.cpp reports on the M5 Max 48GB
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.
- Model not identified · Q4_K_M26.3 t/s
- Qwen3.6 · Q4_K_M124.3 t/s
- Qwen3.5 · Q4_K_M86.5 t/s
- Qwen3.5 · Q4_K_M135.6 t/s
- Qwen3.5 · Q4_K_M111.2 t/s
- Qwen3.5 · Q4_K_M29.2 t/s
- Qwen 3 · Q4_K_M97.7 t/s
- Qwen 3 · Q4_K_M159.7 t/s
- Qwen 3 · Q4_K_M26.4 t/s
- Qwen 3 · Q4_K_M148.5 t/s
- Qwen 3 · Q4_K_M57.4 t/s
- Qwen 3 · Q4_K_M288.6 t/s
- Qwen 3 · Q4_K_M457.1 t/s
- Qwen 2.5 Coder · Q4_K_M106.0 t/s
- Qwen 2.5 Coder · Q4_K_M26.3 t/s
- Qwen 2.5 Coder · Q4_K_M56.0 t/s
- Phi-4 · Q4_K_M56.8 t/s
- Phi-4 · Q4_K_M164.9 t/s
- Phi-4 · Q4_K_M165.0 t/s
- Phi-4 · Q4_K_M56.3 t/s
- Model not identified · Q4_K_M37.2 t/s
- Model not identified · Q4_K_M68.6 t/s
- Model not identified · Q4_K_M107.1 t/s
- Llama 3.2 · Q4_K_M199.9 t/s
- Llama 3.2 · Q4_K_M439.6 t/s
- Llama 3.1 · Q4_K_M101.1 t/s
- Gemma 4 · Q4_K_M110.5 t/s
- Gemma 4 · Q4_K_M173.7 t/s
- Gemma 4 · Q4_K_M25.2 t/s
- Gemma 4 · Q4_K_M116.9 t/s
Frequently asked
How fast is llama.cpp on the M5 Max 48GB?
llamaperf has 38 llama.cpp reports on the M5 Max 48GB, 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.