I would guess it is some combination of: - longer pipelines - more silicon dedicated to speculative execution for more Instruction Level Parallelism - bigger c

I would guess it is some combination of: - longer pipelines - more silicon dedicated to speculative execution for more Instruction Level Parallelism - bigger caches - More analysis of incoming instruction stream to get more ILP and out of order execution But yeah, more clock is more power is more heat.