Speaker: Dr. José Moreira, IBM
Slides: https://hpc.fau.de/files/2025/12/Why-Matrix-Processing.pdf
Abstract:
It took vector processing approximately 30 years to move from the domain of supercomputers to mainstream computing. In comparison, matrix processing hit the ground running and has arguably been mainstream since inception. Today, matrix processing units are pervasive in modern CPUs and GPUs, as well as in special-purpose processors. In this talk, we will review the principles of matrix processing and discuss its inherent advantage over scalar and vector processing. We will also discuss different approaches to implementing matrix processing units at the instruction-set architecture and micro-architecture level, and the applications and relevance of matrix processing in both AI/ML and more traditional HPC applications. We will also go over some of the challenges and limitations of matrix processing and argue that modern processing elements must offer a balance of scalar, vector, and matrix processing.
For a list of past and upcoming NHR PerfLab seminar events, please see: https://hpc.fau.de/research/nhr-perflab-seminar-series/
