LLMCheck: Apple M5 Ultra (512GB), MLX, Llama 3.1 405B Q4_K_M decode
4.0 tok/s (decode, per-request)
Fields
id
benchmarks/llmcheck-m5-ultra-llama31-405b-q4km-mlx-decodeMetric
decode_tok_s
Value
4
Unit
tok/s (decode, per-request)
Model
Llama 3.1 405B
Methodology
THIRD-PARTY MEASURED (LLMCheck open dataset, dated 2026-08). M5 Ultra 512GB, MLX, Llama 3.1 405B Q4_K_M, per-request.
Measured by
third_party
Reproducible
yes
notes
The capacity headline: a 405B dense model runs at all on a desktop chip, at ~4 tok/s (batch-offline speed). Only the 512GB config has the headroom.
Notes
The capacity headline: a 405B dense model runs at all on a desktop chip, at ~4 tok/s (batch-offline speed). Only the 512GB config has the headroom.