Predicting Resource Bottlenecks in HPC Environments Using AI: A Data-Driven Method for Identifying and Visualizing Compute and Memory Limitations via Correlation Analysis
S. Vijaya Shree, M. Ashok · 2025
The use of artificial intelligence (AI), especially correlation analysis and feature influence scoring, shows strong potential for identifying performance bottlenecks in High-Performance Computing (HPC) environments. Yet, the lack of standardized ways to visualize and interpret these bottlenecks limits reliable diagnostics. This paper introduces a new framework that detects memory and compute bottlenecks using AI-based methods without predictive classifiers. It analyzes how hardware and workload features-such as memory cycle speed, bus width, memory channels, overclocking, accelerator bandwidth, and workload type-affect system performance. By generating influence scores and visualizing results through log-scaled bar plots and heatmaps, the framework offers a clear, reproducible approach to identifying bottlenecks. This work improves interpretability in HPC monitoring and sets the stage for scalable, explainable AI tools that optimize workload distribution and resource use in modern HPC systems.