🚀 go-pugleaf

Clifford Chen <cliffc@fujitsu.fai.com> writes:
>Fujitsu Limited today announced that it has developed and will begin
>marketing worldwide a new supercomputer capable of delivering
>unprecedented peak performance of 355 gigaflops (355 billion floating
>point operations per second) with only 222 processors -- a single
>processor performance level ten times greater than any parallel system
>available today.

I know there is little point in arguing with press releases, but the
Cray Y-MP C90 is certainly available today, and has parallel processors,
and offers 1 Gigaflop per processor, so I don't see how 1.6 Gigaflops
per processor can be a 10x improvement.

Not to mention that the CM-5 goes up to 16000 processors and 2000
Gigaflops.  Don't these guys read each other's press releases?  :-)

>The new system, called the VPP500 vector parallel processor, will be
>available for commercial shipment in September of 1993.

Off the top of my head, the price-performance point of this system
doesn't seem very attractive, especially for this time period.  At $122M
for the full system, a number which Fujitsu mysteriously forgot to
mention in the press release, that looks like at best $350K/GF.
(Systems of realistic size would presumably be worse.)  Aren't large TMC
and Intel systems currently around $250K/GF, and available this year?

Of course, peak Gigaflops aren't the measure of a computer system, but
since they are the measure that Fujistu chose to use....

>If Fujitsu's price/performance claims prove out, the product could
>represent a significant competitive challenge to Cray Research's
>flagship C-90," observes Industry Analyst Gary Smaby, president of the
>Smaby Group.

Perhaps Industry Analyst Gary Smaby knows something we don't.  Like what
those price/performance claims are.  What's the point of providing an
endorsement of the claims but not including the claims themselves?

                                        David desJardins