On Sat, 14 Aug 2004 20:53:24 +0000, Mike Andrews wrote:

In (rec.radio.amateur.homebrew), Keith wrote:

Unfortunately, boiling the CFC also distilled it, leaving any
contamination on the chips. The result came to be known as the "black
plague". Because of the "black plague" the "LEM" (Liquid Encapsulated
Module) was replaced by a similar looking (though shorter) "TCM" (Thermal
Conduction Module) which used pistons on the backside of the chips
(increased to 121 chips) to transfer heat (10W per chip, 1200W total) to
the cold-plate and filled with helium. The TCMs were used throught the
'80s and early '90s for the high-end ECL systems.

Yep. I've got a TCM somewhere at home; I need to take a pic of it
for my web page. IIRC, it has 1024 pins, and the largest ZIF socket
I've ever seen.

I think you'll count a lot more than 1024 pins. The ones I worked on int
the '70s had 1800. There were 1280 signal pins (I rember that name
because of the logic tester - the LT1280).

The cold plate on the TCM is about 4.5" square, very
smooth, and designed to go on a larger water-cooled manifold with some
dozens more TCMs, all getting water from a chiller.

Twelve TCMs for each processor, and nine for each channel (for the 3080s).
A system could have upwards of a hundred of these beasts in it. Later
sytems had fewer since the logic density increased. ...which is what
ultmately killed them. The infrastructure was simply too expensive for
the quantity needed.

They're called "wet-frame" machines by a lot of us who dealt with them,
and especially by those of us who had a leak develop somewhere in the
cooling system.

****ers, eh? I was once tasked to design a "leak detector" for exactly
thi sproblem. I brought in the guts out of a toilet (complete with copper
float) and told them to put these under the floor with a switch. Come on!
Detect a leak in a pipe a few hundred feet long passing many gallons per
minute?

Nowadays one TCM's circuitry fits on a single chip, at a meager fraction
of the TCM's dissipation, and is enormously faster, so that (for 9672
hardware, anyway)

Note that the first CMOS machines were slower than those they replaced (in
direct violation of IBM's "prime Directive"). THough they were
significantly cheaper and with more processors (loophole alert).

IBM ships the machine with a bunch of CPUs and
licenses you the code to turn on as many as you're paying for. If you
want more than ship in the base machine, they'll install another dozen
pretty cheap.

"Install"? ;-) They'll enable as many as you want, for as long as you
are paying. We called this "Rent-A-MIP". THere is even a crypto unit in
there if you want to use it.

And the homebrew and design connections:

They're sure not homebrew, and they're more than one designer can do.

Yeah, there were a few designers doing this stuff. ;-)

--
Keith