Wow, that's pretty small improvement from the K8 (2.7-5.6&#37. They can probably get those results from the K8 if they increased the L2 cache to 2-4mb per core.