M2 Max (38-core GPU) — LLM Benchmarks

Measured LLM inference benchmarks for M2 Max (38-core GPU) across all RAM configurations (32 GB, 64 GB, 96 GB). 6 benchmark rows across 3 models. Compare how RAM affects throughput. Real runs, not estimates.

6Benchmark rows

3Models tested

3RAM configurations

153.0Fastest 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.

32 GBM2 Max (38-core GPU, 32 GB)3 rows · 3 models153.0 tok/s peak 64 GBM2 Max (38-core GPU, 64 GB)1 rows · 1 models22.0 tok/s peak 96 GBM2 Max (38-core GPU, 96 GB)2 rows · 2 models46.4 tok/s peak

All benchmark rows — M2 Max (38-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
M2 Max (38-core GPU, 32 GB)	Llama 3.2 1B Instruct	Q4_K - Medium	—	153.0 tok/s	2551.4 tok/s	—	ref
M2 Max (38-core GPU, 96 GB)	Llama 3.1 8B Instruct	Q4_K - Medium	—	46.4 tok/s	484.1 tok/s	—	ref
M2 Max (38-core GPU, 32 GB)	Llama 3.1 8B Instruct	Q4_K - Medium	—	44.7 tok/s	473.7 tok/s	—	ref
M2 Max (38-core GPU, 96 GB)	Qwen 2.5 14B Instruct	Q4_K - Medium	—	25.2 tok/s	252.5 tok/s	—	ref
M2 Max (38-core GPU, 64 GB)	Qwen 2.5 14B Instruct	Q4_K - Medium	—	22.0 tok/s	222.8 tok/s	—	ref
M2 Max (38-core GPU, 32 GB)	Qwen 2.5 14B Instruct	Q4_K - Medium	—	20.6 tok/s	225.6 tok/s	—	ref

Models tested on M2 Max (38-core GPU)

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 →