The weird stuff I keep finding: 19 bit core memory?
jnc at mercury.lcs.mit.edu
Thu Oct 22 19:49:10 CDT 2020
> From: Chris Zach
> My guess is the H215 has two more core fields on it
I uploaded a (crappy - sigh) image of an H215 I have to here:
and it's clearly not symmetrical, but does have a slightly bugger blank space
than yours. It has a label that says "18x8K" (so 16 bits plus byte parity -
for DATOB bus write operations to one byte at a time), which sounds orrect.
> Odd they could fit up to 20 bits, maybe an early ECC (16b+4 ECC?)
No, this way pre-dates ECC memory (the cost of the extra bits - more cores to
string in - was probably more than it was worth), and probably before ECC
disks too (I think the first DEC ECC disk was the RP04).
> I'll bet that it could work in the 11/05 if one of those boards ever
> blow out.
That's my guess too; _but_ there might be some jumpering/etc issues to adjust
the word width. I.e. the PDP-11 parity H215 is 18 bits wide, which is right
for 16-bit words, whereas PDP-10 parity memory would be 36+1 parity bit = 37
bits wide. So there might be 2 banks on one of those cards, and the PDP-11
memory treats them as separate banks, whereas on a -10 they'd be ganged
together in parallel.
(Whether that's all done on the companion driver boards, and the H21x card
would just bring the wiring of the two banks out to the edge connector in
parallel, letting the driver board do what it wants, I don't know - you'd have
to look at the MM11-L engineering drawings/TM.)
The other possibility is that the PDP-10 memory used these boards in pairs.
We do have the MA10 Maint Manual, but in a quick look it looks like it
doesn't use these boards. So maybe a later -10 memory system?
More information about the cctalk