siliconscore: M4 Ultra 192GB, Llama 3.3 70B Q4_K_M, MLX, avg decode
18.0 tok/s
Fields
id
benchmarks/siliconscore-m4ultra-llama-33-70b-q4km-mlx-lm-decodeMetric
decode_tok_s
Value
18
Unit
tok/s
Model
Llama 3.3 70B
Methodology
COMMUNITY-REPORTED via siliconscore.com audit JSON (row verified=false, source_url=llmcheck.net). Apple M4 Ultra (Mac Studio 2025, 192GB unified, ~819 GB/s), runtime MLX, Q4_K_M-class quant. avg_tok_s = mean generation rate; context_tokens not published. Reported TTFT 2.0s.
Measured by
third_party
Reproducible
no
notes
Unverified community row aggregated by siliconscore/llmcheck. 192GB unified M4 Ultra. TTFT 2.0s also reported.
Notes
Unverified community row aggregated by siliconscore/llmcheck. 192GB unified M4 Ultra. TTFT 2.0s also reported.