[HARDWARE] Cost of crunching (was: Mac Questions)

John L. Bass jbass at dmsd.com
Mon Mar 18 12:49:04 EST 2002


George Mogiljansky <mogil_jan at sympatico.ca> writes:

	Forgive my ignorance, but isn't bandwidth associated/proportional at some
	stage with energy consumption? And isn't increased bandwith requirement
	associated with high speed/cable access?

Only if the tecnology is the same. Here it isn't, we have a very old and power
wasteful analog current mode local loop in most cases - and very low power,
high speed RF connection as the other case.  If you look at the energy per bit
on the local loop for DSL vx analog modem, analog is several orders of magnitude
higher. The DC energy while off hook, is lost in the copper pair resistance
to carry the relatively large current thru miles of small gauge wire.

The power per bit in your analog dialup line, is probably one to two orders of
magnitude higher than the aggregate power per bit for the entire rest of the
"bit's life" as it transitions over high speed digital voice lines to the ISP,
and high speed digital packet lines over the internet. 

The same issues are true inside older low speed computers processing Dnet
blocks. The power per bit is lots higher for older large feature CPU's
used just 5 years ago, in comparison to very small feature CPU's used
today at the same or higher frequency. That's why the power efficiency in
Thord's numbers are two orders of magnitude better for the two faster systems
in comparison to the 486 machine at nearly the same power.  The 486 burns
113W to crunch at 0.093 MKeys/s, and the faster machines burn 117W and 115W
to crunch 2.5 and 3.4 Mkeys/s.

The LCIII that started this thread is half the power, and one third the
performance of the 486 Thord tested ... significantly worse power efficiency.
My position is that these marginal machines are wasting huge amounts of
power for almost no benifit - We as the Dnet community should suggest that
they just get turned off.

John
--
To unsubscribe, send 'unsubscribe hardware' to majordomo at lists.distributed.net



More information about the Hardware mailing list