by Spike » Tue Aug 05, 2014 9:47 pm
as qbism says, mdl and md2 both use byte positions with origin+scale
the difference between the two is that md2 has that origin+scale on a per-frame basis while mdl is per-model.
this means that if your model moves/resizes then *EVERY SINGLE VERTEX* must move in order to stay aligned on the byte grid.
md3 is 16bit with 1/64th qu precision (iirc). if you move the verts at the extremity, nothing else needs to move. however, small movements (1/32th - 1/64th qu) will end up visibly snapped to the grid. maybe the user won't notice.
iqm has full 32bit float precision. being a skeletal format, it can be slower to render. frames are expressed as per-bone rotations or translations, as such, just moving verticies around randomly for each frame is a bad idea. you'll need to animate them properly, on a bone-by-bone basis. good luck fixing up problems after the fact. yes, you will need to be more careful with your rigging.
dpm as a format is pointless. its advantage is its obscurity, but that's basically it. use iqm. iirc, exporters for this format are somewhat compatible with halflife's smd stuff, so you may find it easier to export to if that's the you're already familiar with that (not a contradition: toolchains often make or break a format regardless of any technical merits of the format itself) .
.