This is Intel's 80-bit floating point. The reason the size is 128 is that it needs to align on a word boundary.