Resurrecting the CDC 6500 Supercomputer [pdf] | Hacker News Reader

Resurrecting the CDC 6500 Supercomputer [pdf]

ieeexplore.ieee.org

57 pointsKatydid18 comments

I found this part of the article interesting in its "spareness":

> Final assembly was quite challenging because you have to put all of the resistors and diodes in one side and then try to slide them across and into the right hole on the other side. It’s a rather tedious process, but it works.

That description seemed to echo "cordwood construction" for those modules, but I wanted to know if this was actually the case. The best I could find was this image of a CDC 6600 logic module:

https://upload.wikimedia.org/wikipedia/en/thumb/6/64/CDCcord...

...cordwood construction.

From what I understand, such modules allowed for greater part densities (and probably better airflow for cooling), but at the cost of repairability: Since they were made with thru-hole components, soldered on one side, if a component in the middle of the board failed, it was a very time consuming task to unsoldered all the parts on one side to be able to separate the halves to repair the module.

Gah I used to absolutely hate to disassemble those for parts. But they were very cheap because everybody else hated them too and at the local flea market there was a guy selling those 5 for 2 Guilders, hard to resist.

Also, the remaining legs would be super short, on the order of 2-3 mm if you were lucky.

https://www.youtube.com/watch?v=8zk1zByXpBk#t=04m18s

Seymour Cray would sacrifice maintainability for performance every time. Example: The Pogo Pins used on the Cray 2:

https://en.wikipedia.org/wiki/Cray-2#/media/File:Cray-2_modu...

The Cray 2 also had to have the Flourinert drained before repairs could be made.

It’s challenging to keep the memory modules running. They consist of a woven mat of core memory with fine wires running through the cores to read and write the data bit that was stored in each core. It’s difficult to repair the core memory because everything is so small:

Each thing that looks like it might be a core is actually four cores on edge. Each core has five wires going through it and they’re all suspended by the mat of wires that go through the cores.

It's funny to see "everything is so small" -- these are giant compared to lithographically formed semiconductor memory. But it's not exactly wrong either. The parts are small enough to challenge the human sensorium and motor control, yet still large enough to tempt interaction using your own eyes and fingers.

I purchased a... sheet? Block? of 1024 bits of core memory that came from a Soviet computer, and it was hand-woven. I have a hard time picturing people hand-weaving this core memory, and I'm guessing his core memory is smaller than the stuff I've got. Absolutely amazing.

The story was that CDC had the densest cores (5 mil?) made at what was termed the Hong Kong core house.

I'm using this machine remotely sometimes, when it works. I'm incredibly grateful that they provide this service for interested individuals.

What was most fascinating to me was seeing how BASIC was one of the languages available. What was even more interesting was how similar this implementation is to those available on the micros of the 80's like the Commodore 64. I could pretty much just start typing away and it just worked.

In contrast, I still haven't gotten any Fortran code to run.

I'm really pleased that there are people doing this kind of stuff. It's awesome to see.

> The pins were manufactured by AMP, which was still in business. They had taper pins in gold or tin, but not in the silver that was required for the CDC 6500. With a bit of convincing, negotiating, and LCM+L’s promise to purchase 50,000 pins, AMP agreed to produce a limited run of silver pins:

Can anyone say why silver pins are needed?

Silver seems to me to be the worst material for a connector.

Silver-plated contacts are fairly widespread in use, e.g. better XLR connectors, some Fastons, Hartings etc. use them. They are unproblematic with contact designs that scrape the natural oxide layer off. Mixing platings may prevent this mechanism from working properly, resulting in bad contact.

Yea I think because of the softness you couldn't switch to gold only partially or it'd wear exceedingly fast. It sure about other metals though.

Mixing metals can also lead to corrosion. Luckily this seems to be a "spare no expense" rebuild.

> Can anyone say why silver pins are needed?

> Silver seems to me to be the worst material for a connector.

It's pretty good actually.

Low resistance, very good skin effect. That's why it's usually silver over copper, all of the advantages and not nearly the same price.

Most quality RF components and connectors are silver plated.

The oxide is conductive. The gold to silver price ratio used to be ridiculous compared to historical levels probably due to the death of film but its always been the cheap alternative.

I highly recommend a visit to the Living Computer Museum if you find the time. I had a few hours to spare on a Seattle visit last summer and Bruce was happy to show me the CDC 6500 and his other work.

As a CS student at Michigan State in the late 70s I have fond memories of FORTRAN, COBOL, and assembler programming on the 6500.

It was interesting to see many of the design principles from that mid-60s system show up in RISC designs in the 80s and beyond.

My first exposure to a machine bigger than a 6502-based home computer was a Cyber at the University of Lowell (MA). Not sure the model number, ?600. In 1984 the freshman CS classes were in Pascal, on print terminals. I think my year was the last to use that machine; the following year it was replaced by a VAX 11/780 running VMS.

FYI: link opens a pdf

Reply on news.ycombinator.com