Extensive tests have been run against the VAX and IEEE floating point instructions.  There are two minor deviations under Avanti from the behavior of real AlphaServer 400 hardware.

1.With IEEE floating point instructions, the least significant bit of the fraction may differ between Avanti and a real AlphaServer 400.  This is due to a rounding difference.  There should be very few applications where this difference is significant. 

2.After an overflow occurs in a multiplication or division instruction, the result (which is always invalid) differs between Avanti and a real AlphaServer 400.  The overflow condition is properly signaled, so this should not be a problem for a well-behaved application.