Awesome! Love this kind of stuff.
This is back when IBM was light years ahead of everyone else.
Apparently there are 4 capacitors per package? If so, why 4 connections per capacitor?
Small correction to this amazing video:
@24:30
The paper you inserted says the solder alloy is 97Pb/3Sn which is Lead / Tin, not Tin/Copper as you stated.
This also fits better with the softness of the solder.
i wonder if theys ibm computer systems where this cpus where used,also supported that giant hdd what dave teardown in one of his videos?
Small correction to this amazing video:
@24:30
The paper you inserted says the solder alloy is 97Pb/3Sn which is Lead / Tin, not Tin/Copper as you stated.
This also fits better with the softness of the solder.
Don't think so. I strongly doubt that that alloy has such a low melting point that is is basically liquid at room temperature. It has to be something else.
I worked on one of these from ~94 to 99. Machine ended up being sold for scrap. A 2 frame 9121 (model 732 I *think*)
I got a souvenir, a couple of the heat sinks from that machine.
They are about 5 KG each and look like aluminum.
Looks like 97Pb/3Sn has a 315C melting point according to US Patent 7838954
"The use of such high-melting alloys for
chips provides a hierarchy in soldering temperatures so that
the 97Pb-3Sn solders (melting temperature 315° C.) on the
side of the chip do not melt and the under bump metallurgy is
not consumed during the next level ?ip chip assembly using
loW melting solders such as eutectic 37Pb-63Sn solders With
a melting temperature of 183° C."
Cool.
I have been to the factory that made these in the late 1980s. It was manufactured at a site that no longer exists: IBM Manassas Virginia.
I was working for the US Air Force Space Division doing military space electronics and they were one of our major vendors. I visited IBM frequently at multiple sites.
So IBM's big challenge was mastering was getting the ceramic shrinkage exactly characterized - the ceramic is a "multi-level" ceramic that has multi-level interconnects just like an IC's metallization except it's all laid down as multiple wet ceramic clays with metal layered and then fired. The magic is still achieving the exact dimensions despite shrinkage during firing. No one really has duplicated the capability until the last 10 years with stuff like Amkor and TSMC are doing with chiplets.
We were interested in the density, weight and reliability for space applications. It was high density with high performance with low total weight compared to PCBs and other alternatives.
Yes the die are "flip-chips" and simply pressed by the heat sinks and springs holding them in.
The interconnects in inside the ceramic substrate. The die are NOT all the same. Some may be memory/cache, some processors, some glue logic.
Brings back memories!
Why is the CPU assembly so complex for what appears to be a modest number of MIPS even back then? Are the instructions far more complex than x86 instructions or is it more about memory and I/O bandwidth?
There was some other technology IBM produced that was fascinating. The 028 and 129 card punches had a thing in them called a code plate and an elephants trunk for printing characters. The code plate had a large array of tiny square protrusions, where with a certain x and y displacement, and then displacing in the z direction would hit specific protrusions on the elephant's trunk to print a character (5 x 7 dot matrix). I never worked out how these code plates were made. The lynx belt printer was also one of their achievements.
Another fascinating device were the early Winchester hard disks. One I worked on could be as big as 5.1MB in the System/32, but it took two men to lift it. They cost today's equivalent of about $200,000. Let's not forget the 5225 "washing machine" printer. Very high speed dot matrix printer, that had 8 heads firing off at once, with massive transistors driving high current to the wires. Ceramic substrates were common on all this equipment. You will find them on any aluminium covered IBM chip - just peel the cover off and a jelly covered the chips (not an oil).
There were a lot of innovations IBM had that were not used elsewhere, possibly due to patents. I was fortunate to have worked these and a lot more in my first job at IBM.
By the way, the first TCM's around 1980 were about $25K, so in today's money, about $250K. And that was the IBM INTERNAL price as I recall. You would not want to have had one replaced and not be covered by maintenance contract.
By the way, the first TCM's around 1980 were about $25K, so in today's money, about $250K. And that was the IBM INTERNAL price as I recall. You would not want to have had one replaced and not be covered by maintenance contract.
I found an article from the early 90's about IBM hiking the prices on these modules. The exact one in this video was $225,000
refurbished.
Small correction to this amazing video:
@24:30
The paper you inserted says the solder alloy is 97Pb/3Sn which is Lead / Tin, not Tin/Copper as you stated.
This also fits better with the softness of the solder.
Yes, I confused this with standard 97/3 lead free solder.
Any datasheet, or schematic available for this...?
I worked at IBM rival Amdahl for awhile. One reason for "lower than you would expect" MIPS was resiliency.
Not to get into the weeds of logic design too much, but unexpected settings of the state machines in the CPU were monitored and if one was encountered, the CPU would pause and probe the logic levels of surrounding state machines and then decide what was wrong and overwrite a correct values over the miscreant. The CPU would resume as if nothing had happened. There was a time limit of how much incoming data could be buffered without overflowing. A modem would report the hit back to the home office.
New designs or misbehaving CPUs could have hundreds of these soft errors an hour and it would just keep chugging along with uptimes sometimes into the hundreds of days. No outages allowed for big iron customers and the customers would pay crazy prices - as seen in the price of this IBM module, wow.
Nice teardown. But did the IBM guys actually had to bodge wire that monster?
That could probably the reason why not all places are populated. Some places are not working correctly.