LLMCheck: Apple M5 Max (128GB), MLX, Llama 3.3 70B Q4_K_M decode
12.0 tok/s (decode, per-request)
Fields
id
benchmarks/llmcheck-m5-max-llama33-70b-q4km-mlx-decodeMetric
decode_tok_s
Value
12
Unit
tok/s (decode, per-request)
Model
Llama 3.3 70B
Methodology
THIRD-PARTY MEASURED (LLMCheck open dataset). M5 Max 128GB, MLX engine, Llama 3.3 70B Q4_K_M, TTFT 2.4 s, per-request decode.
Measured by
third_party
Reproducible
yes
notes
MLX vs Ollama lands at the same 12 tok/s on this model — decode on a dense 70B at Q4 is pure bandwidth, so the engine difference is noise-level.
Notes
MLX vs Ollama lands at the same 12 tok/s on this model — decode on a dense 70B at Q4 is pure bandwidth, so the engine difference is noise-level.