FINIT checks for unmasked floating-point error conditions before performing the initialization; FNINIT does not.
On the Pentium processor, unlike the Intel387 math coprocessor, FINIT and FNINIT clear the error pointers.