MoE-CAP: Benchmarking Cost, Accuracy and Performance of Sparse Mixture-of-Experts Systems

Yinsicheng Jiang (University of Edinburgh) · Yao Fu (University of Edinburgh) · Yeqi Huang (University of Edinburgh) · Ping Nie (Tencent) · Zhan Lu (University of Edinburgh, University of Edinburgh) · Leyang Xue (University of Edinburgh) · Congjie He (School of Informatics, University of Edinburgh) · Man-Kit Sit (University of Edinburgh, University of Edinburgh) · Jilong Xue (Microsoft Research) · Li Dong (Microsoft Research) · Ziming Miao (Microsoft) · DaYou Du (University of Edinburgh, University of Edinburgh) · Tairan Xu (University of Edinburgh, University of Edinburgh) · Kai Zou (NetMind.AI) · Edoardo Maria Ponti (University of Edinburgh) · Luo Mai (University of Edinburgh, University of Edinburgh)

cap radar diagramdeployment scenarioshardware platformsheterogeneous computememory resourcesmixture-of-expertsmoe-capperformance benchmarkingsparse memory bandwidth utilizationsparse model flops utilizationsparsity-aware performance metricssystem costtrade-offs

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. This benchmark is available on Github: https://github.com/sparse-generative-ai/MoE-CAP.