@emeb Using 230 always-running struck stiff-string models, each with a continuous damper amount obeying soft/sostenuto/open-damper/half-damper, and cross-resonance between those and a soundboard model, and both of those summed through a set of formants representing the shape of the board and position of the lid. I'd expect it to take up most of the DE10-Nano.
Patch parameters would be things like string gauge, which key the undamped range starts at, which are the 1/2/3-string keys (unichords, bichords, trips), hardness of the hammers, softness of the felt, bounce, do we want modeled keybed whack, lid position, cabinet shape/size, etc. If you get crazy you could tune each individual string, so that one piano your uncle has with the horrible F# could be modeled with the right patch setting.
There's no way I could ever produce a sampled piano that's even as good as what you get in your average acoustic piano preset, so I figure if I do acoustic piano at all, it will be modeled, like modartt pianoteq, but on an FPGA.