Jan Ackmann - Reduced Precision Computing for Weather and Climate Models
|Starts:||14:00 29 Nov 2019|
|Ends:||15:00 29 Nov 2019|
|What is it:||Seminar|
|Organiser:||Department of Mathematics|
Jan Ackmann from University of Oxford will be speaking at the seminar.
Weather and Climate (W&C) prediction models are required to satisfy strict time-to-solution and energy-to-solution constraints and thus need to be as computationally efficient as possible on modern supercomputers. To explore possible gains in computational efficiency, we follow various approaches: Precision reduction for floating point operations, exploring alternative number formats, and replacing model components with machine-learned surrogates. Resulting computational savings could then be reinvested where they are needed more urgently.The justification for these approaches – that inevitably introduce additional model errors – is motivated by the presence of irreducible uncertainties in W&C model predictions. These uncertainties are due to the interplay of the W&C models’ two main components, the dynamical core, a discretization of the Navier-Stokes equations, and the so-called model physics – a collection of stochastic parametrizations for the unresolved subgrid-scale processes (turbulence, cloud physics, convection,…). In the presence of the resulting uncertainties, high precision is deemed unnecessary for many computational operations and model components.
The first part of the talk will be about the group’s work on the use of low precision arithmetic for various model components (Spectral dynamical cores, Adjoint calculation, Data Assimilation, Legendre Transforms, and model physics), where often a level of half precision is found feasible. Also, alternative number formats such as Posits (emulated in software) and neural network approaches that replace parts of the model physics are discussed.
The second part of the talk will be a more detailed account on reduced-precision preconditioned elliptic solvers in dynamical cores.
Role: Postdoctoral Research Associate
Organisation: University of Oxford
Travel and Contact Information
Frank Adams 1
Alan Turing Building