How we'll speed-up model

  • Largely concentrate on atmosphere, as ocean at ORCA1 is much faster than atmosphere at N96 (different if we're considering N216 & ORCA025)
  • We'll speed-up different aspects/speed-ups of the model, e.g.
    • Use 'high' optimisation
    • Increase timestep
    • Reduce vertical levels
    • Increase timestep for full UKCA and radiation
    • Put more chemistry fields into steady state (this reduces advection costs)
    • Look to reduce time in solvers
  • To start with, likely to consider each speed-up in isolation on an AMIP run
  • Typically for each speed-up we'll probably consider two levels
    • OMG level, e.g. a 1 hour timestep
    • Moderate level, e.g. a 30 or 45 minute timestep
  • Documentation: any platform should be fine, e.g. UKESM ticket