Mixed precision computing reduces energy for scientific simulations
Mixed-Precision Computing for Scientific Discovery: Formats, Co-Design, and Responsible Approximation
Hardware ArchitectureMathematical Software
Summary
Running large scientific calculations usually takes a lot of computer power and energy. The authors describe how using numbers with different levels of accuracy—sometimes less detailed than usual—can save energy and time. They organize the many ways this can happen, including new computer chip designs and smart software that checks when accuracy is enough. Their idea is to use low precision aggressively but have ways to catch and fix errors, so the results stay trustworthy.
What this means in practice
- •For scientific computing teams: Design simulation workflows that use mixed precision to reduce energy use while maintaining trusted results through escalation strategies.
- •For hardware architects: Develop accelerators that support various numeric formats and co-designed software to exploit mixed precision for energy-efficient computing.
A position paper. It proposes an approach and reports no results.
Authors
Emmanuel Agullo, Hartwig Anzt, Daniel Bauer, David Bindel, Alfredo Buttari, Alexandru Calotoiu, Erin Claire Carson, Pasqua D'Ambra, Ieva Daužickaitė, James W. Demmel, Jack Dongarra, Iain Duff, Massimiliano Fasi, Dominik Göddeke, Stef Graillat, Laslo Hunhold, Roman Iakymchuk, Fabienne Jézéquel, Nils Kohl, Harald Köstler, Jakub Kružík, Julien Langou, Xiaoye Sherry Li, Hatem Ltaief, Piotr Luszczek, Yuxin Ma, Theo Mary, Mantas Mikaitis, Hiroyuki Ootomo, Daniel Osei-Kuffuor, Enrique S. Quintana-Ortí, Ulrich Rüde, Jennifer Scott, John Shalf, Linda Stals, Rasmus Tamstorf, Stefan Turek, Petr Vacek, Bastien Vieublé, Rio Yokota
Abstract
Reduced and mixed precision have moved from a niche optimization to a central design axis in scientific computing and engineering, driven by energy constraints, heterogeneous accelerators, and the convergence of simulation and machine learning. This paper organizes the landscape around seven coupled themes---number formats, floating-point emulation, emerging architectures, hardware/software co-design, relation to other approximations, software design, and precision as a multilevel resource ---and, for each theme, synthesizes the state of the art, future directions, and open questions. We emphasize \emph{energy per trusted solution} as the core objective, and we frame \say{recklessly responsible} computing as a pragmatic doctrine: exploit low precision aggressively, but with systematic detection, escalation, and certification pathways.