cprice@vianet.UUCP (10/02/87)
About the goings-on a Cray Research...
Cray Research has had a couple designers (and design teams) for some time.
Seymour did the Cray-1, and Cray-2.
Steve Chen did design work on the XMP, a multiprocessor (1-4 CPUs)
whose processors are (basically) reimplemented Cray-1s
(same instructions, faster cycle, much smaller physically)
with some extra instructions/features for multiprocessing.
The yet-to-be-sold (as far as I've heard) YMP is also supposed to be
largely Steve Chen's work and is, again, a further refinement of the
existing processor line (XMP) but a cycle time in the low nanoseconds
(like around 5ns) comparable to the Cray-2.
Seymour has been working on the Cray-3,
a gallium arsenide machine much like the Cray-2 but with 16 processors.
The CPUs are supposed to be ~2.5 times as fast as Cray-2 processors.
The MP project (after-the-3) machine, a 64-CPU multiprocessor,
was being worked on by Chen.
The loosely-managed company finally noticed that he was spending a whole
lot more money than they had planned on spending without getting
very close to a real machine. The news reports aren't very specific
but he was apparently doing *really* basic research (the news items
keep mentioning lasers and semiconductor technology).
The company (i.e. "they") decided that the whole thing was going to cost
too much for them to finish and was too risky.
They believe that their successes have been from using non-risky technology
and using that to build a simple, but blindingly fast, machine.
Chen left after the project was canned and is apparently trying to get
a company together to continue his work. According to news reports the
departure was friendly and Cray is willing (under undisclosed conditions)
to let him continue working on this stuff (taking existing work with
him and giving them some percentage of later profits????).
Cray Research has quickly come up with a new game plan in the wake of
the not-so-wonderful publicity. Seymour will begin work on the Cray-4
next year (after turning the mostly-finished Cray-3 over to Lester Davis
to bring to market). The Cray-4 will be a refinement of the Cray-3
architecture with 64 CPU's using faster GaAs. The stated target
for speed is a twice the Cray-3: 1 nanosecond clock and 100 Gflops.
While giving no timetable, Seymour has noted that *his* cycle time
is 3-4 years so we could see this in 92-93.
--
Charlie Price {hao stcvax nbires}!vianet!cprice (303) 440-0700
Western Digital - Boulder Technology Center
2900 Center Green Ct. South / Boulder, CO 80301hwang@uicsgva.UUCP (10/09/87)
Cray let Chen to continue the research of MP on the basis that he may not join other companies. He may start a company of his own to continue the MP project.