Re: Pi to 10 trillion places


Peter Webb wrote:

> This would have to be the stupidest hardware test I have ever heard of.
>
> Doing add, multiply, shift 2 trillion times tells you only very little more
> than doing it a million times, and tells you nothing about 90% of your CPU.
> It would be like testing a new car design by driving it in reverse for 2,000
> miles and pretending that it is a comprehensive test.
>

You say that it's just a simple case of "doing add, multiply shift 2
trillion times". I think that's oversimplifying things. It took 12
people 4 years to get the 1.2 trillion figures last time the record was
beaten. I'm sure these 12 worked on other projects in the meantime, but
it still doesn't seem that it's the trivial exercise you make it out to
be.

I think the motivation for participating in such a project should be
obvious: If tomorrow you built a supercomputer, and you wanted to gain
worldwide exposure, what application would you choose? I agree the
largest QR decomposition ever performed in 24 hours would be a
worthwhile project, but it won't get much attention.

Calculating pi to more places than it's ever been done before is not
going to get you the nobel prize. It will however guarantee you
extensive press coverage. This would have been HItachi's motivation in
getting involved in Kanada's project. Their 1152 processor
supercomputer used in Kanada's last effort occupied no.5 in the list of
the top 500 supercomputers in its heyday, and still hasn't dropped out
of the list today. As I understand it they used 512 of them in Kanada's
effort.

.