The sparse Mixture-of-Experts (MoE) architecture is increasingly favored for scaling Large Language Models (LLMs) efficiently, but it depends on heterogeneous compute and memory resources. These factors jointly affect system Cost, Accuracy, and Performance (CAP), making trade-offs inevitable. Existing benchmarks often fail to capture these trade-offs accurately, complicating practical deployment decisions. To address this, we introduce MoE-CAP, a benchmark specifically designed for MoE systems. Our analysis reveals that achieving an optimal balance across CAP is difficult with current hardware; MoE systems typically optimize two of the three dimensions at the expense of the third-a dynamic we term the MoE-CAP trade-off. To visualize this, we propose the CAP Radar Diagram. We further introduce sparsity-aware performance metrics-Sparse Memory Bandwidth Utilization (S-MBU) and Sparse Model FLOPS Utilization (S-MFU)-to enable accurate performance benchmarking of MoE systems across diverse hardware platforms and deployment scenarios.
View on arXiv@article{jiang2025_2505.11415,
title={ MoE-CAP: Benchmarking Cost, Accuracy and Performance of Sparse Mixture-of-Experts Systems },
author={ Yinsicheng Jiang and Yao Fu and Yeqi Huang and Ping Nie and Zhan Lu and Leyang Xue and Congjie He and Man-Kit Sit and Jilong Xue and Li Dong and Ziming Miao and Dayou Du and Tairan Xu and Kai Zou and Edoardo Ponti and Luo Mai },
journal={arXiv preprint arXiv:2505.11415},
year={ 2025 }
}