Geerling ai-benchmarks: RTX 4090 (in Intel 265K PC), Ollama, DeepSeek-R1-14b (Q4 distill), 92.49 tok/s eval rate, 464W peak
92.49 tok/s eval rate (Ollama-reported decode)
Fields
id
benchmarks/geerling-rtx4090-deepseekr1-14b-q4-ollama-decodeMetric
decode_tok_s
Value
92.49
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 #43, obench.sh on Ollama). Host: Intel Core Ultra 265K custom PC. Model: DeepSeek-R1-14b Ollama tag (Q4_K-class GGUF distill). Eval rate 92.49 tok/s; peak system power 464.2W during the run. Per-request decode, batch 1, Ollama default context. Detailed run log in the linked GitHub issue.
Measured by
third_party
Reproducible
yes
Energy basis
wall power at the plug measured alongside the run; 464.2W peak reported in the table
Mean power (W)
464.2
Power scope
node
notes
Geerling also ran the same model on an RTX 4090 attached to a Pi CM5 (83.31 tok/s, 397.5W peak) — the 10% delta is the Pi's PCIe link and CPU bottleneck, not the GPU. Both figures are single-request decode.
Notes
Geerling also ran the same model on an RTX 4090 attached to a Pi CM5 (83.31 tok/s, 397.5W peak) — the 10% delta is the Pi's PCIe link and CPU bottleneck, not the GPU. Both figures are single-request decode.