Comparative Performance Analysis of Workload on Enterprise GPUs with Consumer Platforms Accelerated by CUDA Graphs

• Leandro L. Retzlaff URI
• Calebe C. Pereira URI
• Helena P. Veltri URI
• Murilo Faganello URI
• William T. Klein URI
• Alexandre S. Roque URI

DOI: https://doi.org/10.5753/semish.2026.20788

Resumo

This work investigates the feasibility of reproducing benchmarks originally run on datacenter GPUs such as the NVIDIA A100 and RTX 8000 using consumer-grade graphics cards, focusing on the NVIDIA GeForce RTX 3050 and GTX 1060 with CUDA Graphs support. Seven NAS Parallel Benchmarks (BT, LU, SP, EP, IS, MG, and CG) are evaluated across problem classes W, A, B, and C. Results show that the RTX 3050 delivers stable performance, typically 6×–12× slower than the A100, while VRAM limitations severely constrain the GTX 1060 for larger-scale problems. Although enterprise GPUs remain essential for massive, memory-bound workloads, modern consumer hardware combined with CUDA Graphs enables economical reproduction of moderate scientific experiments, supporting the democratization of high-performance computing research.

Referências

D. Cortes, C. Juiz, and B. Bermejo. Estudio de la eficiencia en la escalabilidad de gpus para el entrenamiento de inteligencia artificial, 2025. arXiv:2509.03263.

O. Diakun and P. Czarnul. Investigation of cuda graphs performance for selected parallel applications. In International Conference on Computational Science, page 130. Springer, 2025. DOI: 10.1007/978-3-031-97635-3_16.

S. Ghimire et al. Cost-effective deep learning infrastructure with nvidia gpu. Kathmandu University Journal of Science, Engineering and Technology, 2025. DOI: 10.48550/arXiv.2503.11246. arXiv:2503.11246.

S. N. Gowda et al. Watt for what: Rethinking deep learning’s energy-performance relationship, 2023.

Y. Hong and D. Kim. Performance and efficiency gains of npu-based servers over gpus for ai model inference. Systems, 13(9):797, 2025. DOI: 10.3390/systems13090797.

A. Mencattini et al. Evolutionary model merging on consumer gpus, 2025. P. O. A. Navaux and M. da Silva Serpa. Desafios do processamento de alto desempenho.

In Seminário Integrado de Software e Hardware (SEMISH), pages 39–49. SBC, 2021. DOI: 10.5753/semish.2021.15805.

R. K. Shah. Global gpu network (ggn): Performance benchmarking and comparative analysis against traditional gpu computing models, 2024.

M. Špet’ko, O. Vysockỳ, B. Jansík, and L. Říha. Dgx-a100 face to face dgx-2— performance, power and thermal behavior evaluation. Energies, 14(2):376, 2021. DOI: 10.3390/en14020376.

R.-F. Wu, Y. Wang, and D. Kutscher. Affordable hpc: Leveraging small clusters for big data and graph computing, 2024.

H. M. Zangana, F. M. Mustafa, and S. Li. Large language models in cybersecurity, 2024.