On Thu, 4 Dec 2003 lerninlinux@comcast.net wrote:

> I have two identicle boards, Asus A7V.  Both I need to do some testing
> with.  One supposedly has problems with the onboard ide controllers.  I
> am wanting to use an offboard ide controller so that's not the issue. I
> have always seen articles on overclocking, but I have an old AMD 1700,
> that I was wanting to UNDERclock to the boards max (1200 if I remember
> right).  This is some old stuff that has been donated to me for my
> Mythtv project.

I don't know anything about the trends of overclocking culture, but I have 
played with overclocking with interesting results.  Most people tweak the 
motherboard's internal clock either by jumpers, internal settings, or 
during runtime with varying degrees of frequency granularity.  It is 
possible to underclock to zero hertz (dc voltage) where the CPU heartbeat 
is suspended in time...

My adventure started with a digital VCO chip taken from an old Winchester
8" floppy drive.  It was a 74LS629 daul voltage control oscillator.  With
a multiturn potentiometer, a single capacitor, and a 5 volt supply, I
could go from 0 Hz to 50 MHz.  I would remove the motherboard's oscillator
crystal and inject this signal into one of its pins.  This frequency would
then be multiplied by the motherboard's set ratios to determine its
various CPU and bus frequencies.

I could slide the CPU frequency from normal to the point of crashing.  I 
could see what machine code instructions failed first.  I could see race 
conditions, bus contention, overheating problems, etc.

Most interesting were the extremely low frequencies.  Most operating
systems use a clock to interrupt tasks at given intervals to give all
processes a turn and schedule events.  If any interrupts are granted by
the operating system, the CPU will put the current task on the stack and
check out its new assignment.  Yes, the stack will crash.  Often with
interesting results.  This way I scould determine how much overhead an
operating system had and how to minimize it.  I forgot to mention dynamic
memory chips require its capacitor cells to be refreshed at a rapid rate.  
Underclocking is very educational.

Power consumption with some CMOS computers could be reduced to virtually
zero with extremely low frequencies.  Unfortunately, there seems to be too
many transistor circuits on todays chips to really notice a difference due
to the leakage at transistor edges from extremely compact densities.  The
power consumption of CMOS circuitry should be zero at zero hertz as the
transistors are insulated gates.  Increase the frequency and your current
consumption comes from the logic transitions.  Current should increase at
an exponential rate with regard to frequency.

This may be more than what you want to know.  These projects enter into
the domain of semiconductor engineering.  Fun stuff.