Matrix Engines for High Performance Computing:A Paragon of Performance or Grasping at Straws?

TL;DR

This paper critically examines the practical benefits and limitations of matrix engines in modern processors for high performance computing and machine learning, using empirical data and cost-benefit analysis.

Contribution

It provides an in-depth survey and analysis of matrix engines' performance, costs, and potential misuse in HPC and machine learning contexts.

Findings

Empirical data temper enthusiasm for matrix engines in HPC.

Cost-benefit analysis shows limited advantages in certain scenarios.

Highlights potential misuse if matrix engines are assumed to be free.

Abstract

Matrix engines or units, in different forms and affinities, are becoming a reality in modern processors; CPUs and otherwise. The current and dominant algorithmic approach to Deep Learning merits the commercial investments in these units, and deduced from the No.1 benchmark in supercomputing, namely High Performance Linpack, one would expect an awakened enthusiasm by the HPC community, too. Hence, our goal is to identify the practical added benefits for HPC and machine learning applications by having access to matrix engines. For this purpose, we perform an in-depth survey of software stacks, proxy applications and benchmarks, and historical batch job records. We provide a cost-benefit analysis of matrix engines, both asymptotically and in conjunction with state-of-the-art processors. While our empirical data will temper the enthusiasm, we also outline opportunities to misuse these dense matrix-multiplication engines if they come for free.