🚀 go-pugleaf

I've been looking at the IEEE Computer Society magazine for Nov.'92. There
is a photo of what looks like a water-cooled backplane with a bunch of
motherboards plugged into it. There is also the following blurb:

   Fujitsu Ltd. has developed a supercomputer that delivers 355 billion
floating-point operations per sec with 222 processors. The VPP500 vector
parallel processor features a proprietary crossbar network that allows
simultaneous communication to move between individual processing
elements (PE's) at 800 Mbyte/sec with each PE operating at 1.6 Gflops.

   Distributed global memory space reaches a capacity of 55,000 Mbytes.
This allows memory space physically incorporated in each PE to be used
for storage as well. Programmers can thus us traditional programming
methods to develop high speed parallel programs.

   The upgradable LSI system uas fabricated from gallium arsenide and
runs UNIX SVR 4. System pricing begins at approximately $10 million. The
company plans commercial shipment in Sept. 1993.


   This raises alot of questions. Are there any Fujitsuans on the net
who might be able to post archetectural details?

Is this a new instruction-set architecture, or is it compatable with
something else?

How does it do on benchmarks like big linpacks, or preferably, SPECrate?
Is Fujitsu a member of SPEC?

How many chips make up the CPU of one PE? What's the clock rate? How
big is the cache? is main memory DRAM? How fast?

Lots more questions come to mind, but lets see if anybody out there can
answer any of these...

				- Lee Campbell  <elwin@media.mit.edu>