M3 Max Qwen 3 32B Q8_0 prefill (llama.cpp, siliconscore ref)
111.18 tok/s (prompt eval)
Fields
id
benchmarks/m3max-qwen-3-32b-q8-0-llama-cpp-ctx20172-ppMetric
prefill_tok_s
Value
111.18
Unit
tok/s (prompt eval)
Model
Qwen 3 32B
Methodology
Silicon Score benchmarks.csv reference run. M3 Max (GPU count not published, 64 GB): Qwen 3 32B Q8_0 via llama.cpp. context_tokens=20172. Prompt eval tok/s.
Measured by
third_party
Reproducible
yes
Thermal state
unknown
notes
siliconscore reference run; source_url=https://www.reddit.com/r/LocalLLaMA/comments/1kk0ghi/speed_comparison_with_qwen332bq8_0_ollama/.
Notes
siliconscore reference run; source_url=https://www.reddit.com/r/LocalLLaMA/comments/1kk0ghi/speed_comparison_with_qwen332bq8_0_ollama/.