Scalability of the ARM Nvidia Grace Superchip for Deep Learning Applications

• Thiago Araújo UFRGS
• Philippe Navaux UFRGS

DOI: https://doi.org/10.5753/eradrs.2025.6822

Resumo

The advance of Artificial Intelligence (AI) has heightened the demand for computational resources, presenting challenges in scalability and energy efficiency. The low-power ARM architecture is a promising option for AI workloads. This study assesses the scalability of the ARM Nvidia Grace Superchip in Deep Learning (DL) using a model that predicts referrals for severe diabetic retinopathy. We analyze performance in terms of speedup and energy consumption across 1 to 144 cores. Although speedup increases with more threads, the gains are less pronounced at higher core counts due to resource saturation. Energy consumption, however, significantly decreases, showing a 95% reduction when utilizing all cores. These results emphasize the Superchip’s potential for scalable and energy-efficient AI, especially in demanding applications. Future research will focus on memory usage, latency, and distributed learning.

Referências

Banchelli, F., Vinyals-Ylla-Catala, J., Pocurull, J., Clasca, M., Peiro, K., Spiga, F., Garcia-Gasulla, M., and Mantovani, F. (2024). Nvidia grace superchip early evaluation for hpc applications. In Proceedings of the International Conference on High Performance Computing in Asia-Pacific Region Workshops, pages 45–54.

Dash, S. (2025). Green ai: Enhancing sustainability and energy efficiency in ai-integrated enterprise systems. IEEE Access.

Ruhela, A., Cazes, J., McCalpin, J., Del-Castillo-Negrete, C., Li, J., Liu, H., Chen, H., Lu, C.-Y., Milfeld, K., Zhang, W., et al. (2024). Performance analysis of scientific applications on an nvidia grace system. In SC24-W: Workshops of the International Conference for High Performance Computing, Networking, Storage and Analysis, pages 558–566. IEEE.