[Info-vax] The Future of Server Hardware?

[glen herrmannsfeldt]

JF Mezei <jfmezei.spamnot at vaxination.ca> wrote:
(snip)

> The bragging about being green isn't so much about less cooling
> requirement but rather more efficient cooling.  So the bragging doesn't
> include reduction of electrical cost of running a gazillion servers and
> the heat they generates. It focuses or more efficient heat extraction.

> When you have a gazillion servers, you also have a gazillion power
> supplies, and multiple gazillion fans always running in those servers.
> While modern servers have the ability to vary fan speed based on heat in
> CPU, there is still much of a waste with the "idle" machines still
> running fans and still running a power supply.

Ferrite core transformer based switching power supplies are
pretty efficient. 

In the Altair 8800 days it was usual to use a big iron core
transformer, recitify and filter to about 8 volts, then with an
on-board linear regulator go down to 5 volts for TTL. 

About 3/8 of the power is lost in the regulators.

You will notice in the iron transformer based wall plug mounted
power supplies that they get warm, but in the newer ferrite core
switching supplies they pretty much don't. The loss in an iron
transformer with no load is pretty high compared to ferrite.

Stacked servers need more fans to get the air through than
a deskside or desktop that is otherwise open. Some have temperature
sensitive fans, and maybe switching regulators for the speed control.

Until recently, the power used by CMOS was pretty much proportional
to the transistor switch frequency. Quantum tunneling through the
ever thinner gate oxide is now a significant power user.

-- glen