We analyse the physical properties of 121 SNR ≥ 5 sub-millimetre galaxies (SMGs) from the STUDIES 450-μm survey. We model their UV-to-radio spectral energy distributions using magphys+photo-z and compare the results to similar modelling of 850-μm-selected SMG sample from AS2UDS, to understand the fundamental physical differences between the two populations at the observed depths. The redshift distribution of the 450-μm sample has a median of z = 1.85 ± 0.12 and can be described by strong evolution of the far-infrared luminosity function. The fainter 450-μm sample has ∼ 14 times higher space density than the brighter 850-μm sample at z ≲ 2, and a comparable space density at z = 2–3, before rapidly declining, suggesting LIRGs are the main obscured population at z ∼ 1–2, while ULIRGs dominate at higher redshifts. We construct rest-frame ∼ 180-μm-selected and dust-mass-matched samples at z = 1–2 and z = 3–4 from the 450-μm and 850-μm samples, respectively, to probe the evolution of a uniform sample of galaxies spanning the cosmic noon era. Using far-infrared luminosity, dust masses and an optically-thick dust model, we suggest that higher-redshift sources have higher dust densities due to inferred dust continuum sizes which are roughly half of those for the lower-redshift population at a given dust mass, leading to higher dust attenuation. We track the evolution in the cosmic dust mass density and suggest that the dust content of galaxies is governed by a combination of both the variation of gas content and dust destruction timescale.