31 Oct
2010
31 Oct
'10
5:39 p.m.
Microdata core stacks had these huge resistors in the inhibit circuit
that would literally set the board and system on fire.
Panel mode, R register 0xA010 in the switchers (write) and run, wait for
smoke or PS failure.
Microcoded machines with easy access to this sort of 1/2 cycle write
would hit the board every 200us, when the requirement was for 600us for
a 1/2 cycle read, and 1000us for a Write. The writes would not complete
that fast, but the 1600 (and 800's) logic was such that you would cycle
the board fast enough to cause the damage.
I know that individual cores would simply stop functioning over time. I
worked with a company in Santa Ana, Memtec, which repaired the Microdata
and a few other local minicomputer companies core stacks, but mostly
Microdata's because the principles in the company were ex Microdata
field service.
I recall seeing one with the stack open and there was a small spot in
the middle of the stack where they had clipped out the failing core from
the three wires, extended (or rethreaded) the replacement back into the
stack, and it stood maybe 1/2mm to 1/4mm out from the array. They had a
dab of hotmelt or something on the rework since there were uninsulated
welds where they did the splicing. (mind you very small welds).
A friend acquired their working microscope in the early 80's for a
keepsake, and it was a large Nikon with binocular high eye relief eyepieces.
Microdata had two people who did corestack repair, and some of the
stacks made in Puerto Rico came with repairs, especially in the 16k
arrays. Microdata manufactured their own 16k core stack, but bought
outside for the 8K's. I think the original 4k stacks they used were
made by Dataram, I could look if anyone is interested.
The 4k stacks were probably dated from 1968 or so, and had a large metal
plate over the array which was about 8" by 8".
anyone have any idea about whether this was mechanical, or was it
possibly for magnetic reasons in early core stacks?
The 8K arrays and 16K arrays (as well as the Ampex / CDC arrays) were
all just circuit boards with a grid etch pattern covering them no full
metal plate.
Jim
On 10/31/2010 4:44 PM, Chuck Guzis wrote:
On 31 Oct 2010 at 19:12, Charles Dickman wrote:
So how does a single core fail? A crack I guess.
Maybe overheating
that could change the characteristics of the core materials.
That was my
understanding--and the reason that core was in an oil
bath to start with.
I vaguely recall something related about the CDC 7600, where it was
possible to write pathological code that would repeatedly hit the
same bank of core (core, even on the PPUs was interleaved, but it was
very fast--for the time--27.5nsec core) and cause said core to
overheat and the machine to throw parity errors. The solution was to
integrate the frequency of bank accesses and slow the system down if
they became too frequent. I don't recall how it was done, however.
--Chuck