HPL-MxP Benchmark: Mixed-Precision Algorithms, Iterative Refinement, and Scalable Data Generation

2025-09-23

AI Summary

Existing benchmarks struggle to simultaneously evaluate AI accelerator performance and ensure numerical reliability. To address this, we propose HPL-MxP—the first mixed-precision linear algebra benchmark designed for exascale supercomputing. Our method integrates low-precision LU factorization with non-stationary GMRES iterative refinement, eliminating fixed-iteration constraints; introduces diagonal scaling to enhance numerical stability for ill-conditioned problems; and features a scalable, large-scale matrix generator supporting terabyte-scale inputs. Experimental results demonstrate that HPL-MxP achieves exascale-level throughput on mainstream AI accelerators while maintaining controlled backward error. Thus, HPL-MxP establishes a novel evaluation standard for HPC–AI convergence—offering scalability, robustness, and hardware adaptability.

Technology Category

• Machine Learning: Hardware-aware ML
• Humans and AI: Other Foundations of Human Computation & AI
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Economics, Online Markets and Human Computation: Research challenges in human and human-AI computation
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

We present a mixed-precision benchmark called HPL-MxP that uses both a lower-precision LU factorization with a non-stationary iterative refinement based on GMRES. We evaluate the numerical stability of one of the methods of generating the input matrix in a scalable fashion and show how the diagonal scaling affects the solution quality in terms of the backward-error. Some of the performance results at large scale supercomputing installations produced Exascale-level compute throughput numbers thus proving the viability of the proposed benchmark for evaluating such machines. We also present the potential of the benchmark to continue increasing its use with proliferation of hardware accelerators for AI workloads whose reliable evaluation continues to pose a particular challenge for the users.

Problem

• Evaluating mixed-precision algorithms for high-performance computing benchmarks.
• Assessing scalable data generation methods and their numerical stability.
• Benchmarking exascale systems and hardware accelerators for AI workloads.

Innovation

• Mixed-precision LU factorization with iterative refinement
• Scalable data generation for numerical stability evaluation
• Exascale-level compute throughput on supercomputing hardware