Good thing about BD is that they are not sitting still but working on improving it. Both IPC and power/clock have much room to grow. This is brand new design on a brand new process. They didn't meet the estimated performance (check my post in SoC topic) and I guess they are working hard to fix whatever is bottlenecking the design. IF they meet their 10-15% "faster" estimates(big if) on a yearly basis,they will do fine in server space. Note that the 10-15% faster is just core+clock improvements. They plan to add more "modules" to both server and desktop so this should be also taken in consideration. By the 2014, they will be running 20nm and packing 14 modules for a total of 28 threads per chip while the core+clock improvements are estimated at 33-55%. So 75% more "threads" and 33% (worst case) more "speed" per core => 1.75x1.33=2.33x performance in 3 years timeframe. We should also remember that the BD generation that comes in 2014 is supposed to carry on board GPU for GPGPU coprocessing capability (GCN based probably).