Related Experiment Video
Updated: Sep 4, 2025

Fabrication, Operation and Flow Visualization in Surface-acoustic-wave-driven Acoustic-counterflow Microfluidics
Published on: August 27, 2013
Fluid Simulations Accelerated With 16 Bits: Approaching 4x Speedup on A64FX by Squeezing ShallowWaters.jl Into
Milan Klöwer1, Sam Hatfield2, Matteo Croci3
1Atmospheric, Oceanic and Planetary Physics University of Oxford Oxford UK.
This study demonstrates that using 16-bit floating-point (Float16) arithmetic on the Fugaku supercomputer can significantly accelerate Earth-system simulations. Researchers achieved speedups up to 3.8x by optimizing fluid circulation models for lower precision, making computations faster and more efficient.
Area of Science:
- Computational science
- Climate modeling
- High-performance computing
Background:
- Earth-system simulations typically use 64-bit double precision (Float64), which may be excessive given uncertainties.
- The Fugaku supercomputer, based on A64FX microprocessors, supports 16-bit low-precision (Float16) arithmetic.
Purpose of the Study:
- To investigate the performance of 16-bit arithmetic (Float16) for Earth-system simulations on the A64FX architecture.
- To develop and evaluate techniques for addressing precision and dynamic range limitations in Float16.
Main Methods:
- Developed ShallowWaters.jl, a fluid circulation model using entirely 16-bit arithmetic.
- Implemented compensated summation for precision-critical time integration.
- Created the Sherlogs.jl analysis-number format for logging simulation results.
- Systematically rescaled equations to fit the limited Float16 range (6 × 10⁻⁵ to 65,504).
Main Results:
- Achieved speedups of up to 3.8x on A64FX using Float16.
- Incorporating compensated time integration yielded speedups up to 3.6x.
- Optimized ShallowWaters.jl to utilize 97% of representable Float16 numbers.
Conclusions:
- 16-bit arithmetic is a viable and competitive method for accelerating Earth-system simulations on modern hardware.
- Techniques for managing precision and range issues enable efficient use of Float16 in complex models.
- Findings suggest potential for significant computational gains in climate modeling and related fields.
More Related Videos
08:49Impacts of Free-falling Spheres on a Deep Liquid Pool with Altered Fluid and Impactor Surface Conditions
Published on: February 17, 2019
11:14A Microfluidic System with Surface Patterning for Investigating Cavitation Bubble(s)–Cell Interaction and the Resultant Bioeffects at the Single-cell Level
Published on: January 10, 2017
Related Concept Videos
Accelerating Fluids
The motion of the liquid within this infinitesimal cylinder is considered to obtain the pressure difference. Three vertical forces act on this liquid:
Uniform Depth Channel Flow: Problem Solving
Uniform Depth Channel Flow
Pressure of Fluids
Newtonian Fluid: Problem Solving
A velocity gradient forms within the fluid when a Newtonian fluid is placed between two parallel plates, with...
Surface Tension of Fluid
Surface tension varies...