在消费级显卡上实测9个开源大模型的能耗，发现模型架构和量化策略比参数量更影响能效。
Energy Efficiency of Locally Deployed LLMs: A Preliminary Quantitative GPU Power Benchmark on Consumer Hardware
- 用RTX 4060Ti测试9个1B~7B参数模型的实时功耗，每2秒采样一次
- gemma3:1b和llama3.2:1b能效最优，每词元仅0.56焦耳，吞吐超170词元/秒
- 7B-Mistral能耗是高效模型的4.4倍，qwen3.5:2b因推理延长导致每提示能耗异常高
本地部署大语言模型因隐私与本地推理需求日益流行，但消费级硬件上的能耗成本尚未充分评估，现有基准多聚焦于准确率。本文在单张消费级显卡RTX 4060Ti 16GB上，对九个开源LLM（1B至7B参数）进行了可复现的硬件级能耗基准测试。使用Ollama推理引擎，通过nvidia-smi以2Hz频率采样GPU功耗，固定提示集下评估平均/峰值功耗、每提示总能耗（J/prompt）、每输出词元能耗（J/token）及吞吐量（tok/s）。结果表明，除参数量外，模型架构与量化策略显著影响能效。其中gemma3:1b与llama3.2:1b表现最佳，每词元能耗低至0.56 J/token与0.65 J/token，吞吐量超过170 tok/s；而7B-Mistral模型每词元能耗最高达高效模型的4.4倍。值得注意的是，qwen3.5:2b因内部推理过程延长，导致每提示能耗异常偏高，凸显效率评估中需区分词元生成模式。
原文摘要 · Abstract (English)
The local deployment of large language models (LLMs) is gaining traction due to privacy concerns and the desire for on-premise inference. However, the energy costs on consumer hardware remain poorly characterized, as most benchmarks focus solely on accuracy. This paper presents a reproducible, hardware-level energy benchmark of nine open-source LLMs (1B to 7B parameters) executed on a single consumer GPU (RTX 4060Ti 16GB). Using the Ollama inference engine, GPU power draw was sampled at 2Hz via nvidia-smi across a fixed prompt set. We evaluate mean/peak power, total energy per prompt (J/prompt), energy per output token (J/token), and throughput (tok/s). Our findings suggest that factors beyond raw parameter count, including model architecture and quantization strategy, drive energy efficiency. Specifically, gemma3:1b and llama3.2:1b achieve the lowest energy cost (0.56 J/token and 0.65 J/token) and the highest throughput (>170 tok/s). In contrast, the 7B-Mistral model consumes up to 4.4x more energy per token than the most efficient model. Notably, qwen3.5:2b exhibits anomalously high per-prompt energy due to extended internal reasoning, highlighting the need to distinguish between token generation modes in efficiency metrics.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。