ICL Research Profile

Overview

With the rapidly expanding landscape of mixed- and multi-precision methods, the ongoing cross-pollination between high-performance computing (HPC) and machine learning (ML) or generative artificial intelligence (AI) is leading to intelligent computational steering of large-scale simulations. As these disparate scientific fields share the hardware platforms, exploiting their wide range of computational modes has led to proliferation of multiple representations of floating-point data. Taking full advantage of them is this effort’s main goal.

Against the aforementioned backdrop, ICL’s high-performance libraries (also those produced by internet-scale companies, hardware vendors, national laboratories, and academic institutions) spearhead the recent algorithmic progress in exploiting multiple precisions for increased efficiency in achieved performance, required communication, or optimized storage needs. The techniques used in this effort employ floating-point representations such as limited precision, quantized integers, and modular precision ecosystems, among others. Note that the lossless or lossy compression approaches can independently benefit HPC codes as their algorithmic and accuracy advances are developed in parallel to the mixed-precision aspects.