Seafloor roughness varies considerably across the world's ocean basins and is fundamental to controlling the circulation and mixing of heat in the ocean and dissipating eddy kinetic energy. Models derived from analyses of active mid-ocean ridges suggest that ocean floor roughness depends on seafloor spreading rates, with rougher basement forming below a half-spreading rate threshold of 30-35 mm yr(-1) (refs 4, 5), as well as on the local interaction of mid-ocean ridges with mantle plumes or cold-spots. Here we present a global analysis of marine gravity-derived roughness, sediment thickness, seafloor isochrons and palaeo-spreading rates of Cretaceous to Cenozoic ridge flanks. Our analysis reveals that, after eliminating effects related to spreading rate and sediment thickness, residual roughness anomalies of 5-20 mGal remain over large swaths of ocean floor. We found that the roughness as a function of palaeo-spreading directions and isochron orientations indicates that most of the observed excess roughness is not related to spreading obliquity, as this effect is restricted to relatively rare occurrences of very high obliquity angles (>45 degrees ). Cretaceous Atlantic ocean floor, formed over mantle previously overlain by the Pangaea supercontinent, displays anomalously low roughness away from mantle plumes and is independent of spreading rates. We attribute this observation to a sub-Pangaean supercontinental mantle temperature anomaly leading to slightly thicker than normal Late Jurassic and Cretaceous Atlantic crust, reduced brittle fracturing and smoother basement relief. In contrast, ocean crust formed above Pacific superswells, probably reflecting metasomatized lithosphere underlain by mantle at only slightly elevated temperatures, is not associated with basement roughness anomalies. These results highlight a fundamental difference in the nature of large-scale mantle upwellings below supercontinents and superoceans, and their impact on oceanic crustal accretion.