Using a new microcanonical algorithm efficiently vectorized on a Cray XMP, we reach a simulation speed of
1.5 nsec per update of one spin, three times faster than the best previous method known to us.