25 Nov
2014
25 Nov
'14
6:42 p.m.
On 11/25/2014 02:18 PM, Eric Smith wrote:
Well, the KL10 was not an early CPU design. There are ways
to make good switching
supplies, DEC should have done the engineering. I'm sure
even **I** could have built
a workable switching supply.
Jon
On Mon, Nov 24, 2014 at 8:46 PM, Jon Elson <elson
at pico-systems.com> wrote:
There may have been several versions. The KL10B in a
Decsystem 2020 we had had
one big regulator that was a ring of "christmas tree"
segments with a fan about 10"
in diameter. This was all 30 years ago, so my memory may
have faded just a
little, but I'm pretty sure that was how it was done. Quite
insane. DEC started
using switching regulators in all large DC supplies in the
late 60's, and if ever
there was a machine that cried out for a switching regulator
it sure was the
KL10! So, the decision to use a linear regulator with a
huge voltage drop
must have been some misguided belief that ECL was too noise
sensitive
for a switching regulator. I have built smaller systems
with ECL, and found
it to be quite noise-tolerant.
The KL10B had an immense 3-phase transformer,
star rectifier and capacitor
bank,
About 3.9F worth of capacitors, which was a heck of a lot back in the mid
1970s.
which produced about 12 V DC. This was fed to a
massive linear
regulator on a huge round heat sink.
Actually to 13 massive linear regulators on
multiple large heat sinks. Compuserve ran
a LOT of KL10B's, and designed
their own switching power supplies to save power and air conditioning costs. They
never got DEC to pick up on that, so had to retrofit all the machines as they came
in.
When the KL10 was designed, DEC was already using switching power
supplies in some machines (e.g., PDP-11/40, /45, /55, /70). However,
the load regulation, ripple, and noise specs weren't great, and ECL is
sensitive to the ripple and noise. By the time Compuserve designed
their replacement, switching power supply technology had improved
somewhat.