Quote Originally Posted by [XC] flat-four View Post
I seriously doubt that AMD would spend all this time developing a new processor and would gain this little.
I and several other people have reviewed the list of actual changes to the K10 integer units and I came up with a prediction of 5% win over K8 per-core same-clock. I have documented this prediction here and at 2cpu and you won't see my posts haven't been edited since then.

The new L3 cache won't do much for per-core performance since L2 is already big enough for most real-world applications common demands (it was in K8) and going through an additional level of cache has a cost associated to it.

The point about K10 is to delivery many-core low-power systems with good IPC in the server segment and quad-core low-cost systems in the desktop and home use segment.

The huge and unexpected win that Core2 had over K8 is from my observation largely based in the L2 cache design with it's aggressive and speculative prefetch. AMD did not try to do this for K10. The L3 cache in K10 is complicated and it will be useful to improve finely threaded code running in all 4 cores of a L3 cache. But the K10 L3 cache does not include improvements that would be useful per-core like Core2 introduced.