3 Apr
2013
3 Apr
'13
2:26 p.m.
On Wed, 3 Apr 2013, Pontus wrote:
I apparently made the false assumption that we both
shared the view on
what the Road Runner really is, a collection of CPU's each in its own
motherboard each in its own case. So I carelessly thought that when I
referred to the possibility of replacing the CPU it would be clear that
each CPU would be replaced from it's socket.
I seems that you view Roadrunner as _one_ computer and then it certainly
becomes ridiculous to talk about _the_ CPU in _a_ socket and I
understand why you think I was off the wall.
Yeah, that's most of the confusion. I honestly don't know anything about
Roadrunner, nor anything about the machine being dismantled after 4 years.
Since "top supercomputer" is CERTAINLY going to be upgraded (or replaced)
it seemed that upgradeability and/or expansion would be an extremely high
priority in the design.
I'm sorry for the confusion and I hope you are not
just being sarcastic
with me, that translates badly in text.
No, no sarcasm at all.
I know that I don't know anything about high end machines, but I was
shocked that the design life was only 4 years.
Going back to topic. It certainly is possible to
design a super computer
that is modular. The SGI Origin2k is a wonderful example of that, though
it only scales to a certain limit (128 CPUs I think). Heck, even the
Roadrunner is quite modular, you could probably cut it up into a hundred
different pieces and each piece would be quite a competent cluster of
computer for a researcher willing to pay the power bill.
I like the idea of that, but honestly have no idea of the practicality.
Could multiple 128s be "clustered"?
ANY finite limit will eventually get in the way.
--
Grumpy Ol' Fred cisin at xenosoft.com