Marc's pages
Marc's home
My documentation
Presentations
UKESM core (2017-)
2018
11 December
5/6 November
26 September
17 May
12 April
Aims of UKESM-fast
How we'll speed up model
Properties of speed-up
Where does hybrid fit in?
Measuring speed-up
Compiling with 'high'
Increase timestep
Reducing vertical levels
Calling UKCA & radiation less often
Advecting fewer chemistry fields
Reduce time in chemistry solver
Reduce time in aerosol solver
Reduce nwbins for UKCA_ABDULRAZZAK_GHAN
Any more speed-ups?
Concurrent threading and sharing 2nd thread
Hybrid latest
H2O feedback
Hybrid test jobs
Reduce vertical levels
Background: CMIP5 runs were on about 38 levels. Dan Copsey is running N1280 on 70 levels (he says it's fairly easy to create the ancils now).
Levels
OMG: 38 levels?
Moderate: 70?
Who: ?