A Methodology for Selecting and Ranking High-Performance GPU Configurations in Deep Learning Training

Abstract This study introduces a methodology for the systematic evaluation and classification of GPU configurations used in Deep Learning training tasks. The proposed approach follows a three-stage process that includes the identification of relevant performance variables, the grouping of systems through clustering techniques, and a comparative ranking of the resulting categories based on scalability and efficiency criteria. By structuring the analysis around metrics such as training time, speedup, throughput, and resource efficiency, the methodology enables the discovery of hardware configurations that achieve favourable trade-offs between computational performance and accelerator utilization. Although demonstrated on a specific training workload, the approach is general and can be applied to any dataset or benchmarking scenario that exposes equivalent performance indicators. The results offer a reproducible framework for selecting high-performance GPU setups in large-scale AI environments, contributing to more efficient and cost-effective training infrastructures.