Anika Hennig | Searcharxiv

arXiv2026

Integrating Energy Efficiency into Software Development: Developer Perspectives and Requirements

In the context of the growing energy footprint of information and communication technology, industry optimization efforts have primarily focused on hardware, while the impact of software on energy consumption is often overlooked. Although technical approaches for optimizing software energy consumption have been developed in research, their adoption in everyday development practice remains limited. This study investigates how software developers perceive energy efficiency in their daily work and which requirements and barriers they formulate for AI-assisted tools supporting energy-aware development. As part of the European GreenCode project, ten semi-structured interviews with professional software developers were conducted and analyzed using qualitative content analysis following Mayrings methodology. The identified requirements were subsequently partly interpreted through the lens of the Technology Acceptance Model. The results indicate that energy efficiency rarely plays an explicit role in daily development activities. Instead, energy savings are typically achieved indirectly through performance optimization. Identified barriers to the explicit consideration of energy efficiency include limited awareness and a strong focus on timely delivery. The interviews further revealed requirements for practical tool support, such as actionable optimization suggestions and seamless integration into common development environments. Furthermore, the acceptance of AI-assisted optimization tools strongly depends on transparency regarding the use of data, the actual energy savings compared to the energy consumption of the tool itself, and the disclosure of the training data used. This study contributes a developer-centered perspective on requirements for energy-aware software development tools and provides insights for designing AI-assisted solutions that align with real-world development practices.

cs.SE↗

arXiv2026

Energy Efficiency of Locally Deployed LLMs: A Preliminary Quantitative GPU Power Benchmark on Consumer Hardware

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.

cs.AI↗