← All benchmarks

M4 Pro (20-core GPU) — LLM Benchmarks

Measured LLM inference benchmarks for M4 Pro (20-core GPU) across all RAM configurations (24 GB, 48 GB, 64 GB). 9 benchmark rows across 3 models. Compare how RAM affects throughput. Real runs, not estimates.

9Benchmark rows
3Models tested
3RAM configurations
119.3Fastest avg tok/s

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

All benchmark rows — M4 Pro (20-core GPU)

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

Chip (RAM)ModelQuantRAM req.Avg tok/sPrompt tok/sRuntimeSource
M4 Pro (20-core GPU, 24 GB)Llama 3.2 1B InstructQ4_K - Medium119.2 tok/s2128.8 tok/sref
M4 Pro (20-core GPU, 48 GB)Llama 3.2 1B InstructQ4_K - Medium118.9 tok/s2134.1 tok/sref
M4 Pro (20-core GPU, 64 GB)Llama 3.2 1B InstructQ4_K - Medium118.6 tok/s2145.1 tok/sref
M4 Pro (20-core GPU, 64 GB)Llama 3.1 8B InstructQ4_K - Medium32.9 tok/s349.4 tok/sref
M4 Pro (20-core GPU, 48 GB)Llama 3.1 8B InstructQ4_K - Medium32.7 tok/s361.9 tok/sref
M4 Pro (20-core GPU, 24 GB)Llama 3.1 8B InstructQ4_K - Medium32.5 tok/s359.6 tok/sref
M4 Pro (20-core GPU, 64 GB)Qwen 2.5 14B InstructQ4_K - Medium18.0 tok/s183.0 tok/sref
M4 Pro (20-core GPU, 24 GB)Qwen 2.5 14B InstructQ4_K - Medium18.0 tok/s190.4 tok/sref
M4 Pro (20-core GPU, 48 GB)Qwen 2.5 14B InstructQ4_K - Medium18.0 tok/s189.8 tok/sref

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 →