From: user5857@newsgrouper.org.invalid   
      
   Stefan Monnier  posted:   
      
   > The notion that some paths are critical (and those determine the   
   > frequency that can be used) implies that there are paths which are not   
   > critical, because they're much shorter, e.g. the low-bits of an addition   
   > don't need to wait for the carry-propagation, so they're typically ready   
   > long before the end of the cycle.   
   >   
   > Do CPUs try and save power by making those paths slower (using lower   
   > voltage, or slower and less power-hungry transistors)?   
      
   I admit to using minimum sided transistors in the logic units (slowing   
   it down). But those things rarely consumed enough power to be worth   
   "much" effort.   
      
   On the other hand, there is an entire 'art" to transistor sizing which   
   makes an 11-gate integer adder take only 8-FO4 gates of delay, thus   
   fitting in a ½-cycles allowing for input muxing (forward) and output   
   muxing (result drive) in a combined single cycle::   
      
        |<----------1 cycle---------->|   
        | inMux | integer ADD |outMux |   
      
   {{ this can also be timed as::   
      
        |<----------1 cycle---------->|   
        | integer ADD |outMux | inMux |   
   or   
        |<----------1 cycle---------->|   
        |outMux | inMux | integer ADD |   
      
   I have participated in designs using all three of these timings}}   
      
   So, on a small scale I agree that you can slow down a few paths.   
   Many of them in the control paths of the machine rather than on   
   the data-path.   
      
   On the other hand: slowing down interconnect wires {that are "data   
   path height" long or longer} is seldom profitable, especially as   
   one aims at higher frequencies.   
      
   A different thing happens in  the exponent section of FMAC.   
   You don't make it slower per-sé, instead you allow things to progress   
   without spending the effort to make them more concurrent and thus faster.   
   The Multiplier tree and Baugh-Walley adder take most of the time, so the   
   exponent section only has to be "that fast".   
   >   
   >   
   >         Stefan   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)