评测苹果M4芯片上大模型推理的能耗与碳足迹，发现能效比数据中心高30-40倍
GreenBench: Benchmarking Energy Efficiency and Carbon Footprint of Open-Source LLM Inference on Apple Silicon
- 用真实功耗测量+精准计时，量化苹果M4芯片运行大模型的能耗表现
- 小模型（3-3.8B）比大模型（7-9B）每字节能耗低62%，吞吐高2.6-4.2倍
- 推荐Qwen 2.5（7B）平衡性能与能效，Llama 3.2（3B）适合低延迟场景
大型语言模型（LLMs）推理的环境影响日益受到关注。尽管绿色人工智能研究聚焦于数据中心GPU和嵌入式平台，但搭载统一内存架构的苹果M4 Pro芯片上的大模型推理能效仍缺乏研究。本文提出GreenBench基准框架，评估五款开源大模型（3-9B参数）在三个NLP任务上于配备48 GB统一内存的M4 Pro芯片上的能效、吞吐量及碳足迹。通过macOS powermetrics进行直接功耗测量，结合Ollama的纳秒级时间精度，发现持续推理期间CPU+GPU封装功耗仅0.47 W，系统总功耗为8-12 W，单用户部署下每令牌能效比数据中心GPU高30-40倍。较小模型（3-3.8B）每令牌能耗较大型模型（7-9B）降低62%，吞吐量提升2.6-4.2倍。帕累托分析显示，Qwen 2.5（7B）在57% MMLU准确率下实现59 tokens/s的最优能效权衡；而Llama 3.2（3B）在175 tokens/s下适合低延迟应用。本文提供封装级与系统级每令牌能耗，并给出印度与美国电网的碳排放估算。
原文摘要 · Abstract (English)
The rapid proliferation of Large Language Models (LLMs) has raised concerns about their environmental impact during inference. While Green AI research has focused on datacenter GPUs and embedded platforms, the energy profile of LLM inference on Apple Silicon, with its unified memory architecture, remains unstudied. This paper presents GreenBench, a benchmarking framework that evaluates the energy efficiency, throughput, and carbon footprint of five open-source LLMs (3-9B parameters) across three NLP tasks on an Apple M4 Pro with 48 GB unified memory. Using macOS powermetrics for direct power measurement and Ollama's nanosecond-precision timing, we find that the M4 Pro draws only 0.47 W of CPU+GPU package power during sustained inference, with total system power of 8-12 W, achieving 30-40x better energy efficiency per token than datacenter GPUs in single-user deployment. Smaller models (3-3.8B) deliver 2.6-4.2x higher throughput and up to 62% less energy per token than larger models (7-9B). Pareto analysis identifies Qwen 2.5 (7B) as the optimal accuracy-efficiency trade-off at 57% MMLU and 59 tokens/s, while Llama 3.2 (3B) suits latency-critical applications at 175 tokens/s. We provide per-token energy at package and system levels with CO2 estimates for India and US grids.
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。