AMD: Barcelona quad-core 50% faster than Intel’s quad-core Xeon

[gOJDO]

@carfax
the benchmarks are specFP_rate2006 and specINT_rate2006 on a 2P servers.
Because of the ODMC and the point-to-point buses, the 2P K8 performance scales much better than 2P C2D.
The dualcore K8 is 2.5% faster than C2D in specINT_rate2006, at same frequency, while it is 15.5% faster in specFP_rate2006.
If we consider AMD's bold claims as true, then the 20% advantage over C2D in specINT_rate2006 will become 17% advantage over K8, while the 50% in specFP_rate2006 will become 29% advantage over K8. In average the K10 is 23% faster than K8.

But, these benchmarks are on 2P systems, where K8 shines and scales much better in performance than C2D, because it has an ODMC and a point-to-point buses.
On the desktop, the K8 has no advantage over C2D(no core-FSB-northbridge-FSB-core communication) by having a point-to-point bus. The 1P K8 system has no additional bandwidth provided by NUMA.
The average C2D performance advantage is 20%, compared to a same clocked K8. According to AMD's specXXX_rate2006 estimations, clock for clock, K10 will be 2.5% faster than C2D in average.

@savantu
Stop dreaming. IMC + CSI can't improve performance for 40-50%
For example, K7(Barton with FSB400) vs K8. K8 is a tweaked K7 core with ODMC and new point to point bus. It is only 10% faster clock for clock, due to architectural improvements.