M3 Max (40-core GPU) — LLM Benchmarks

Measured LLM inference benchmarks for M3 Max (40-core GPU) across all RAM configurations (48 GB, 64 GB, 128 GB). 8 benchmark rows across 4 models. Compare how RAM affects throughput. Real runs, not estimates.

8Benchmark rows

4Models tested

3RAM configurations

149.0Fastest avg tok/s

RAM configurations

Each configuration differs only in unified memory. More RAM = larger models fit. Throughput is similar across RAM tiers at the same model size.

48 GBM3 Max (40-core GPU, 48 GB)2 rows · 2 models149.0 tok/s peak 64 GBM3 Max (40-core GPU, 64 GB)3 rows · 3 models107.0 tok/s peak 128 GBM3 Max (40-core GPU, 128 GB)3 rows · 3 models146.3 tok/s peak

All benchmark rows — M3 Max (40-core GPU)

Sorted by avg tok/s descending. Click source badge to see original measurement.

Chip (RAM)	Model	Quant	RAM req.	Avg tok/s	Prompt tok/s	Runtime	Source
M3 Max (40-core GPU, 48 GB)	Llama 3.2 1B Instruct	Q4_K - Medium	—	149.0 tok/s	3399.1 tok/s	—	ref
M3 Max (40-core GPU, 128 GB)	Llama 3.2 1B Instruct	Q4_K - Medium	—	146.3 tok/s	3291.9 tok/s	—	ref
M3 Max (40-core GPU, 64 GB)	Llama 3.2 1B Instruct	Q4_K - Medium	—	107.0 tok/s	2521.6 tok/s	—	ref
M3 Max (40-core GPU, 48 GB)	Llama 2 7B	Q4_0	3.6 GB	65.8 tok/s	691.0 tok/s	llama.cpp	ref
M3 Max (40-core GPU, 128 GB)	Llama 3.1 8B Instruct	Q4_K - Medium	—	45.8 tok/s	587.1 tok/s	—	ref
M3 Max (40-core GPU, 128 GB)	Qwen 2.5 14B Instruct	Q4_K - Medium	—	25.5 tok/s	302.4 tok/s	—	ref
M3 Max (40-core GPU, 64 GB)	Llama 3.1 8B Instruct	Q4_K - Medium	—	25.4 tok/s	377.0 tok/s	—	ref
M3 Max (40-core GPU, 64 GB)	Qwen 2.5 14B Instruct	Q4_K - Medium	—	13.8 tok/s	200.8 tok/s	—	ref

Models tested on M3 Max (40-core GPU)

Llama 2 7B Llama 3.1 8B Instruct Llama 3.2 1B Instruct Qwen 2.5 14B Instruct

Data

benchmarks.json — full dataset · chips.json — chip summaries · benchmarks.csv — CSV export

Data sourced from factory lab measurements and community reference runs. See all chips →