What Artificial Intelligence can do for High-Performance Computing systems?
Pierrick Pochelu, Hyacinthe Cartiaux, Julien Schleich
TL;DR
This paper surveys AI techniques applied to high-performance computing (HPC) with a focus on AI-for-HPC rather than HPC-for-AI. It classifies 74 recent AI-for-HPC papers into six application areas—performance estimation, performance optimization, scheduling, ML surrogates, fault detection, and language-model-based automation—and highlights integration opportunities and cross-cutting challenges. A key finding is that scheduling is the most active area, while performance estimation underpins scheduling, optimization, and automation efforts; GNNs and time-series methods underpin anomaly detection, and domain-specific LLMs aid HPC scripting and automation. The authors argue for integrating AI components via an LLM-based operating system (LLM-OS) but emphasize substantial hurdles in MLOps, standardization, benchmarking, and security. Overall, the work maps a path toward a unified, intelligent HPC ecosystem, while calling for standardized datasets, interoperable APIs, and rigorous benchmarks to realize practical benefits.
Abstract
High-performance computing (HPC) centers consume substantial power, incurring environmental and operational costs. This review assesses how artificial intelligence (AI), including machine learning (ML) and optimization, improves the efficiency of operational HPC systems. Approximately 1,800 publications from 2019 to 2025 were manually screened using predefined inclusion/exclusion criteria; 74 "AI for HPC" papers were retained and grouped into six application areas: performance estimation, performance optimization, scheduling, surrogate modeling, fault detection, and language-model-based automation. Scheduling is the most active area, spanning research-oriented reinforcement-learning schedulers to production-friendly hybrids that combine ML with heuristics. Supervised performance estimation is foundational for both scheduling and optimization. Graph neural networks and time-series models strengthen anomaly detection by capturing spatio-temporal dependencies in production telemetry. Domain-specialized language models for HPC can outperform general-purpose LLMs on targeted coding and automation tasks. Together, these findings highlight integration opportunities such as LLM-based operating-system concepts and underscore the need for advances in MLOps, standardization of AI components, and benchmarking methodology.