For many decades now, HPC served scientific community in a variety of ways that advanced the computational simulation capability based on models that not so long ago remained theoretical. Measuring the progress made by the high-end hardware has become its own field with a vibrant community of performance engineers exchanging the practical experiences and even building theory of performance models that allow the users to gain a deeper comprehension of the multi-million dollar machines. With transition to data science and great demand for deep models for statistical inference, the need for accurate measurement of hardware capability has remained strong. With comparative performance studies that inform the decision makers ranging from purchasing to software implementation and deployment, this talk will offer a glimpse of how scientifically important numerical methods can serve to give insightful information into the potential and the drawbacks of the modern hardware platforms. The talk will feature description of algorithms, historical perspective on the HPC hardware, and results from a variety of systems from relevant benchmarks.

Numerical methods and benchmarking across scales, precisions, and hardware platforms

ACS Fall 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.