[Cryptography] 6600 memory size (was: cryptography Digest, Vol 32, Issue 31)

[Peter Capek]

Jerry Leichter wrote:

Story from years back:  The CDC 6600 - the supercomputer of the early
> 1970's - could be purchased with either (I think) 128KW (Kilo-Words - a
> word was 60 bits) or 256KW of main memory.  NYU purchased one.  They asked
> for the 128KW version.  CDC tried to convince them that they should really
> get the full 256KW.  But the institution had a grant and couldn't swing the
> extra money - probably a couple of hundred thousand in those days.  So CDC
> finally delivered a 128KW version.  But ... the developers who played
> around with the machine found that they could actually get at memory beyond
> the 128KW limit.  Writes and reads to it ... worked just fine.  In fact,
> what they eventually determined was that the machine they had came equipped
> with the full 256KW - CDC had just made a special patch to their copy of
> the OS to have it limit itself to the lower half.  A source within CDC
> eventually explained to them that CDC listed a 128KW option, but that they
> didn't really expect anyone to buy one: If you were spending the tens of
> millions one of these things cost, you were not going to go short on main
> memory.  NYU was, in fact, the first customer to buy one of the "small
> memory" configurations - and CDC hadn't actually worked out the necessary
> manufacturing changes to build one.  So they just shipped the configuration
> they had and patched the OS.
>
> (During the anti-Vietnam-war protests of the late 1960's, a group took
> over the machine room in which the CDC 6600 was stored.  They held it
> hostage - and when they left, the left behind some incendiary devices on
> long fuses.  The machine was barely saved from destruction by some faculty
> members - the story was recently retold.  What I heard from people there
> was that they would not have minded so much if the machine had been
> torched:  Insurance would have replaced it, but the replacement would not
> have been one of the very early runs - I think serial number 4 - which was
> too early to support Extended Core Storage, which used slower but cheaper
> memory - we're talking magnetic core in those days - as a kind of I/O
> device.  The fact the machine was such an early model makes the story about
> main memory more believable.)
>
>                                                        -- Jerry
>

It's not very important, but this recounting is quite wrong, and I'd like
to correct it before it propagates.  I was a system programmer on this
machine for about its first 5 years and knew it well.  The machine had the
maximum size memory, 128K, which we often thought of in decimal terms,
131,072 words, or 131K.   It could not have had more, primarily for reasons
of physical packaging.   The memory was packaged on the same chasses as the
logic, and those chasses had little room to spare.   The CPU's core memory
took up more than a quarter of the space in in the machine, and for reasons
of cable length limits and physical access, it had to be in one frame.   It
is true that the NYU machine was serial #4 - one of about 11 machines built
in Chippewa Falls - and unable to support ECS, extended core.   There's
absolutely no truth to the comments about memory size limitation patches to
the operating system.  There was, though, a 64K version of the machine
available, but if (my!) memory serves, only for later machines not built in
Chippewa, but rather, built in a Minneapolis suburb.  It's maybe also worth
saying that it was a mid-60s machine, rather than early 70s; the NYU
machine was installed in June of 1965.

            Peter Capek
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.metzdowd.com/pipermail/cryptography/attachments/20151231/3184374d/attachment.html>