Using 24 months of 1-min radiometric measurements conducted at two remote arid sites in Kuwait, the impact of sensor technology (thermopile vs. photodiode with rotating shadowband) on the magnitude of the three components of solar irradiance (global, direct and diffuse) is analyzed. The deviations (photodiode minus thermopile) are typically affected by both sun zenith angle and irradiance magnitude. For the global and direct components, most deviations (91% in the case of GHI, 87-91% in the case of DNI, depending on site) are within +/- 5%, and can thus be considered satisfactory. Larger deviations in direct and global irradiance are typically found under low zenith angles (summer conditions). The main source of concern is the negative bias and intricate pattern found in the diffuse deviations, most of the time. Only 46-61% of the deviations (depending on site) are within +/- 5%. The diffuse issue seems to be caused by an insufficient spectral correction of the diffuse reading. The so-called "cat ear" angular issue (a sudden spike in sensitivity for an angle of incidence approximate to 80 degrees, combined with a sudden drop in sensitivity beyond approximate to 85 degrees) is also still present in the direct and global irradiance measurements under clear conditions. The present results underline the imperfect nature of the empirical corrections typically applied to photodiode instruments to improve their irradiance estimates. Nonetheless, the deviations observed here are sufficiently low in general to guarantee good resource assessments, even under harsh and variable desert conditions, to the condition that the photodiode instruments are properly calibrated on site during periods whose atmospheric conditions are representative of the whole year, and their readings are duly corrected with the best possible algorithms.