Jonathan de Boyne Pollard wrote to Will Honea on 03-21-1998
JP> The FIC PA 2012 uses a switching voltage regulator.  I'm told that
JP> that is  good, although I have yet to find an explanation as to why. 
On the one hand, it's a simple matter of physics.  A linear regulator
works by controlling voltage at it's output.  Ignoring internal losses,
the input current is the same as the ouput current and regulation is
effected by dissapating power.  Simplistically, the pwoer dissapated by
the regulator is (input voltage - output voltage) * input current. 
Practically, this means that the current surges in the low voltage
circuits of the processor are passed on to the main power supply as
just that- current surges of the same amplitude.
A switcher works by doing a power conversion - the power drawn from
the input supply and the power drawn from it's output are (ideally)
equal.
Since the CPU (and other low voltage circuits) can produce current
surges of several AMPS, this is an important distinction.  For a simple
example, consider a 5 volt input, 2.5 volt output.  Assume a peak
current of 5 amps on the 2.5 volt side - conservative for a Pentium
class device - and you get 12.5 watts dissapated by the linear
regulator with a 5 amp load on the main supply (25 watts).  With the
switcher, you get almost no dissapation in the regulator and only a
12.5 watt load on the main supply.
Of equally great import, the main supply with a linear regulator is
subjected, in the example, to twice the dynamic load change compared to
the switcher model.  This requires a drastically higher demand on the
main supply with regard to transient response - which translates to
noise potential on the supply lines for the rest of the machine.
The numbers don't appear to be significant, but when you consider real
world conditions and levels it adds up really fast - my soldering
station uses about 15 watts, for example.  It's just a lot of heat and
a lot of high level transients that are avoided by the switcher.  Of
course, this is an idealization;  it ignores the 'hash' generated by
the switching circuits and real regulator efficiencies but the overall
situation is valid.
I looked at the PA2012 but settled for the PA2007 because of the ISA
and memory configuration.  Too many old parts to jump too far all at
once :-{  I have been very impressed, although I've got one nagging
problem.  I thought that any real parity ( x36 ) SIMM would work as ECC
memory.  The board functions quite well with ECC enabled, but I lose
the blasted floppy controller if I enable ECC.  Freezes the system
solid if I access the floppy with ECC enabled!  Curious.....
Will Honea 
--- Maximus/2 2.02
---------------