Old core memory system.

Paul Koning paulkoning at comcast.net
Sun May 6 13:19:28 CDT 2018

> On May 5, 2018, at 2:32 PM, Chuck Guzis via cctalk <cctalk at classiccmp.org> wrote:
> On 05/05/2018 10:23 AM, Pete Lancashire via cctalk wrote:
>> Core temp was a big issue even in commercial environments. You didn't see
>> it temp but you would see core [driver] current.
> The early IBM 7000 series (7070, 7080, 7090) kept core in a
> temperature-regulated oil bath.  Later versions used pre-heated air
> (e.g. 7094 core).
> On the CDC 7600, hitting the same area of care repeatedly could cause it
> to overheat and throw parity errors.   Circuitry to detect this would
> slow-down repeated accesses.
> That was for CM.  I seem to recall someone telling me that there was no
> such provision in PP core and a "jump to self" was sufficient to throw
> an error--but that may be a shaggy-dog story.

The IBM 1620 had a core heater; I remember having to wait for core to warm up after turning the machine on before it would start.

One of the old PDP-11 diagnostics is the "core heating test".  The description said it would hammer a region of memory (physical area) to get it to warm up, to see if there was enough margin for memory to remain reliable.

On CDC memory: The 6000 series PPs always access memory every cycle.  Whenever it doesn'thave anything better to do, the control logic uses the P register (program counter) for address.  So there is no way in a PP to make memory work any harder; it's always a cycle every microsecond (memory running flat out) no matter what.  I'm not positive the 7000 series works the same, but it would seem plausible that it does.

CM, on the other hand, is referenced only when requested.  So CM is normally working less than PP memory.

Parity?  I know the 170 series had parity, didn't think the 7600 did.  The 6000 series does not (except for ECS).  "Parity is for farmers" -- Seymour Cray.


More information about the cctech mailing list