Geerling ai-benchmarks: Mac Studio M3 Ultra 512GB, Ollama, DeepSeek-R1-14b (Q4 distill), 51.85 tok/s eval rate, 227W peak
51.85 tok/s eval rate (Ollama-reported decode)
Fields
id
benchmarks/geerling-m3ultra-deepseekr1-14b-q4-ollama-decodeMetric
decode_tok_s
Value
51.85
Unit
tok/s eval rate (Ollama-reported decode)
Model
deepseek-ai/DeepSeek-R1-Distill-Qwen-14B (Ollama default Q4)
Methodology
THIRD-PARTY MEASURED (Jeff Geerling, geerlingguy/ai-benchmarks issue #42, obench.sh on Ollama). Mac Studio M3 Ultra 512GB. DeepSeek-R1-14b Ollama tag. Eval rate 51.85 tok/s; peak system power 227W. Per-request decode, batch 1.
Measured by
third_party
Reproducible
yes
Energy basis
wall power at the plug measured alongside the run; 227W peak reported in the table
Mean power (W)
227
Power scope
node
notes
Fastest Apple Silicon entry in Geerling's DeepSeek-R1-14b table, but still ~44% slower than the RTX 4090 figure (92.49 tok/s) — the 800GB/s-vs-1TB/s bandwidth gap plus Ollama's CUDA path advantage. Power figure is node-scope wall draw; clocks were not locked (no -lgc equivalent on this Ollama run).
Notes
Fastest Apple Silicon entry in Geerling's DeepSeek-R1-14b table, but still ~44% slower than the RTX 4090 figure (92.49 tok/s) — the 800GB/s-vs-1TB/s bandwidth gap plus Ollama's CUDA path advantage. Power figure is node-scope wall draw; clocks were not locked (no -lgc equivalent on this Ollama run).