Note that with 4+1 -> 4 shader transition they gain almost 25 % increase in perf/mm˛, so they can get 25 % more performance with roughly the same transistor budget. The thing is, are they going to take the performance or shrink the chip.