'long double' is 12 bytes on i386 and 16 bytes on amd64. There is a gcc-specific __float128 type: https://stackoverflow.com/questions/13516476/long-double-gcc-specific-and-float128 Yuri