Just shaved 40% off unsteady HEC-RAS 6.4 2D runtimes on a 12 km reach by switching from uniform 2 m cells to a shear-triggered variable mesh (2–8 m) and keeping structures tight. If you target a Courant around 0.8 and stop chasing tiny timesteps, then fix mass balance first and let roughness calibration do the lifting, peak stage stayed within ±3 cm for me. Anyone else getting better ROI from shear/gradient-based refinement than blanket downscaling?
I’ve had good luck initializing long runs with a 20-minute steady warm-up from the same inflow so the floodplain isn’t “dry”; it knocks down early oscillations and lets me hold CFL near 0.8 without chasing tiny steps. Small caveat: if you’ve got SA/2D connections, keep the steady and unsteady loss settings aligned or you’ll get a mass blip in the first hour.
Biggest ‘free’ speedup I see is I/O: write to a local SSD and push the ‘2D Output Interval’ to 10–15 min (or toggle Mapping off) while calibrating — cuts 20–30% for me with no change in peaks; the disk is usually the slowest guy on the team. Keep the high-frequency outputs for a single final run; how big are your HDF5s?
Quick example: on a 6.4 run with a 2–8 m variable mesh, nudging the 2D “Minimum Depth for Wetting/Drying” up to about 0.05 m (and Face Min Depth about 0.03 m) killed the early edge chatter so I could hold your “Courant around 0.8” without shrinking dt. It shaved about 15% off runtime on a 12 km reach with no change in peak stage, but watch for losing tiny puddles on flats.