I wonder where that 100 Petaflops figure comes from. I suppose that does not mean IEEE-754 compliant floating-point operations, just crude approximations of some sort.