Do I have old software, or does the RS/6000 really mangle its real*16 infinities? For example, inf*1 gives Nan(!) using IBM's "doubled" precision. I guess Tera's real*16 (doubled precision as well) won't match IBM's. Bob