Presumably it's parallelizable. Double the clockspeed, 16x the cores... that'll get you 32x improvement right there, but you've also got way more L3 cache, which might help considerably.