Greenbench: Benchmarking Energy Efficiency and Carbon Footprint of Open-Source Llm Inference on Apple Silicon

Abstract

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 $30-40 \times$ better energy efficiency per token than datacenter GPUs in single-user deployment. Smaller models (3-3.8B) deliver $\mathbf{2. 6 - 4. 2} \times$ 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 tradeoff 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 $\mathbf{C O}_{\mathbf{2}}$ estimates for India and US grids. Taken together, these results show Apple Silicon delivering a $\mathbf{3 0}-\mathbf{4 0} \times$ per-token energy advantage over datacenter GPUs in single-user deployment, positioning consumer hardware as a credible platform for sustainable LLM inference and arguing for energy-per-token to join accuracy and throughput as a standard benchmarking metric.