Mixing Precisions in Numerical Algorithms - HPC Perspective with Modern Hardware Context

Explore the evolving landscape of floating-point arithmetic and mixed precision algorithms in this NHR PerfLab Seminar talk by Piotr Luszczek from MIT Lincoln Lab and The University of Tennessee Knoxville. Delve into the impact of deep learning on data formats and the subsequent changes in hardware implementations. Examine the development of new mixed precision algorithms and their analysis methods in response to these shifts. Gain insights into a range of numerical methods, their convergence properties, and performance considerations in HPC benchmarking. Learn about the speaker's contributions to this emerging field and understand the implications for scientific computing and high-performance computing applications.