4 Jan
2022
4 Jan
'22
10:26 a.m.
On 1/4/22 7:40 AM, Grant Taylor via cctalk wrote:
I guess I thought that since Seymour left CDC to form
Cray Research,
that meant that he was more of an employee at CDC and had less influence
on how it operated as a company.? I would have assumed that someone that
was a founder would have had more influence and tried to improve things
before splitting off and forming yet another new company.
My recollection was over the CDC 8600 project--there is a preliminary
document over at bitsavers:
http://bitsavers.org/pdf/cdc/8600/Preliminary8600RefMan_Aug72.pdf
Sort of a 4-processor 64-bit 7600, but with some important differences,
such as discarding the notion of A and B registers and relying on 16
64-bit registers for everything. P-relative branches and calls and
out-of-user-space "library" routine use.
Apparently, the sharp pencil people were unwilling to fund the thing
fully and Seymour left (with the blessings of CDC which owned stock in
Cray Research) to do his own thing, along with selection of a few key
CDC personnel to help staffing. Afterwards, Cray was forbidden to
recruit from CDC ranks for a number of years.
At the time, the super-duper-computer funding was being funneled into
Jim Thornton's STAR-100 project. You could see a definite difference of
philosophy--the STAR was a virtual-memory vector machine (eventually
became the ETA-10) with scalar subset facilities, whereas the 8600 was a
real-memory fast scalar architecture with no vector facilities. In the
end, the Cray-1 wound up with some vector instructions and registers.
Funding a system like the Cray I or STAR (or any other supercomputer of
the day) required a huge cash commitment and CDC was unwilling to fund
*two* such projects.
--Chuck