M2 (10-core GPU) — benchmark record

M2 (10-core GPU) local LLM benchmarks across 8 GB, 16 GB, 24 GB RAM tiers on Apple Silicon Mac. 7 published rows across 3 models with explicit evidence state and RAM-tier comparison. Peak published speed is 56.5 tok/s.

7Benchmark rows

3Models tested

3RAM configurations

56.5Fastest 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.

8 GBM2 (10-core GPU, 8 GB)1 rows · 1 models56.5 tok/s peak 16 GBM2 (10-core GPU, 16 GB)3 rows · 3 models55.6 tok/s peak 24 GBM2 (10-core GPU, 24 GB)3 rows · 3 models54.8 tok/s peak

All benchmark rows — M2 (10-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 (10-core GPU, 8 GB)	Llama 3.2 1B	Q4_K - Medium	—	56.5 tok/s	804.8 tok/s	llamafile	ref
M2 (10-core GPU, 16 GB)	Llama 3.2 1B	Q4_K - Medium	—	55.6 tok/s	801.8 tok/s	llamafile	ref
M2 (10-core GPU, 24 GB)	Llama 3.2 1B	Q4_K - Medium	—	54.8 tok/s	813.7 tok/s	llamafile	ref
M2 (10-core GPU, 16 GB)	Llama 3.1 8B	Q4_K - Medium	—	14.7 tok/s	140.7 tok/s	llamafile	ref
M2 (10-core GPU, 24 GB)	Llama 3.1 8B	Q4_K - Medium	—	14.7 tok/s	141.7 tok/s	llamafile	ref
M2 (10-core GPU, 16 GB)	Qwen 2.5 14B	Q4_K - Medium	—	8.1 tok/s	74.8 tok/s	llamafile	ref
M2 (10-core GPU, 24 GB)	Qwen 2.5 14B	Q4_K - Medium	—	7.3 tok/s	70.3 tok/s	llamafile	ref

Models tested on M2 (10-core GPU)

Llama 3.1 8B Llama 3.2 1B Qwen 2.5 14B

Data

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

Data from in-house lab measurements plus community-published benchmarks. See all chip families →