E2: Can Multicore Integration Justify the Increased Cost of Process Scalling?

In the last few years processors have hit the power wall and the GHz race is effectively over. Process scaling is yielding little more value than higher integration, providing more die space for extra components. At the same time, the design and manufacturing costs for more advanced processes have increased dramatically. The primary performance advantage of new process nodes has turned into integrating more and more CPU cores each generation. Yet integration issues, on-chip connectivity challenges, off-chip bandwidth demands, testing and software conspire to make this multi-core direction an expensive one in terms of design effort, time to market and platform challenges. How long will the benefits of core count expansion outweigh the costs?