IBM 9121 Thermal Conduction Module (TCM) Teardown

[intabits]

Thermal conduction module from IBM 9121 mainframe processor:-


* TCM weighs 2.2Kg
* Contains up to 121 chips about 8-10mm square
* Each chip has a spring-loaded Copper piston to remove heat
* Up to 10W dissipation per chip
* Up to 600W dissipation per TCM
* Ceramic substrate has:-
   * 63 layers
   * Up to 400m of wiring
   * Up to 2 million (!) vias
* 5Kg air-cooled heatsink to remove heat from TCM

I've had these things for over 20 years, and had a page about them long ago (the TCMs are at the end):-
http://members.optusnet.com.au/intaretro/Packages.htm

Recently I've made a video of them:-


I recently sold one of the TCMs, which is why I made the video, while it was still complete

[dexters_lab]

nice, thanks for sharing. I have been after a MCM (Multi Chip Module) for my unusual IC collection. You have a nice one there!

i've attached a picture of a similar arrangement in an IBM 3084 which is part of Jim Austin's collection. It's older than your system and is water cooled.

[intabits]

Nice that. I've known about them for decades, but never actually seen what it looks like. A lot more tubing than I expected.
I've pulled apart 1401, 7040, 2040s, 3115, 3031s, 4341, 4381,  and the 9121 way back then.  Would like to have gotten my spanners on a 3081 though...

[nad007007]

nice video.

[warrjon]

Wow, this takes me back, last IBM CPU I worked on was the 9021. I have a 3090 TCM in a box somewhere.

[mrz80]

Whoa, TCMs!  When I was a console jockey at NASA/Goddard back in the early 80s we had a 3081. The water cooling was closed-loop, with a heat exchanger that was fed from the building chilled water supply. The supply loop had this huge, gnarly old filter on it that had clogged years ago but was never replaced; the cutout valves were set to bypass the filter. One day we got a new facilities services type who was doing a walk-through of our building. He noticed the water filter in bypass, thought, "That's not right!" and reset the valves to put it back in line. Yeah, you can see where this is going, right?    First thing we knew about it in Operations was about 20 minutes later when the MVS VM threw a whole slew of bright red console errors, the VM console froze, and the machine went into emergency thermal shutdown.  A couple of TCMs in one of the processors had cooked. IBM had to fly down some new ones from New York to get us back up. Epic times in mainframe land! 

[German_EE]

ARRGGGHHHH!!!

I hate this sort of thing. When you bypass something you lock it and you tag it to a) tell others WHAT you did and b) WHY you did it. Fortunately this was not a safety critical error and nobody died.

[katastic]

Old thread but this deserves a mention.

ALWAYS. OBEY. LOCKOUT. TAGOUT.

My dad had to literally save a mans life because of improper lockout tagout when the electrician at the factory thought power was off in a crawl space and began arcing with >40KV electricity shooting through him. My dad found the breaker, turned the power off, pulled his charred body out and began CPR while screaming at others to call 911. He broke a bunch of the mans ribs giving CPR, but the man survived. When the guy woke up in the hospital he just kept repeating, "You saved my life, Mike. You saved my life, Mike!"

There's no excuse to not leave a simple lockout when you "hack" something into being counter-intuitive. Being lazy doesn't just cost dollars, it costs lives. You don't want to be responsible for someone's death, and you don't want to be on the receiving end of it either.