Some new features not known previously/or not well known:
-Two SSE/logical shuffle units
-Using DirectPath instead of VectorPath instructions to take advantage of 128-bit FP units for PSHUFD and SHUFPx instructions(4x faster on those 2 instructions and 2x faster on overall shuffle instructions)
-x87 and scalar FP performance hasn't changed, only vector SSE
(So for PC, its the era of SSE, no more x87 FP)
Similar feature will be on Penryn to improve shuffle performance, named Super Shuffle Engine.
It doesn't look like K10 will outperform Conroe much at all. The bandwidth advantage AMD has is proven to be irrevelent for PC.