Home // International Journal On Advances in Software, volume 19, numbers 1 and 2, 2026 // View article
Authors:
Miren Illarramendi Rezabal
Joseba Andoni Agirre
Aitor Picatoste
Juan Ignacio Igartua
Keywords: LLMs; GenIA; GreenComputing; Code Generation; Energy Consumption; Sustainability.
Abstract:
This research presents a comparative analysis of various Large Language Models (LLMs) for code generation tasks executed without dedicated Graphics Processing Units (GPUs), focusing exclusively on CPU-based environments. While GPUs are typical for large-scale inference, many users rely on CPUs due to resource constraints. However, high-performance chips like the Apple M3 Max introduce a new paradigm for local execution. This study addresses the gap in understanding how LLMs perform across different portable hardware tiers, focusing on the trade-offs between model logic and hardware power. The study evaluates key performance metrics including inference time, code generation accuracy, CPU and memory usage, and energy consumption. To assess the influence of hardware architecture on model efficiency, experiments were conducted across two distinct hardware tiers: a resource-constrained Intel x86 CPU platform and a high-performance ARM-based system powered by the Apple M3 Max. By performing repeated experiments under controlled conditions, we analyze the impact of model size, architectural optimization, and hardware characteristics on computational efficiency. Energy consumption is measured using tools such as CodeCarbon, enabling a detailed examination of the environmental footprint associated with CPU-based LLM execution. The comparison between traditional x86 architectures and modern unified memory ARM systems provides insights into the trade-offs between model precision, computational resource usage, and energy efficiency. The findings contribute practical guidance for developers and researchers seeking to balance performance, sustainability, and hardware constraints in low-resource or portable computing environments, highlighting how architectural differences influence both operational efficiency and environmental impact.
Pages: 82 to 95
Copyright: Copyright (c) to authors, 2026. Used with permission.
Publication date: June 30, 2026
Published in: journal
ISSN: 1942-2628