[9fans] zettaflops workshop

[9fans] zettaflops workshop

[ron minnich]

Some interesting preso's here:
http://www.zettaflops.org/fec07/agenda.html
More to come, I hope.

two tongue-in-cheek quotes from the workshop:

"Supernova physics: there is a lot of interest in this recently
because the fate of the universe depends on it"

and
"It's only recently that they have gotten supernovas to explode at all"

Anyway, really doing some of these simulations well is projected to
require 10^18 FLOPS.

Projection is a machine that has 2^27 CPUs ... since the Great Clock
Rate Climb is now at an end for a while, the only way to scale up is
more parallelism. One vendor contemplates memory addressabilty across
that machine, so, guess what, we just ran out of 64-bit address space,
since 2^37 bytes per CPU is not enough.

Life is getting interesting again, since you can no longer count on
clock rate increases to cover for bloatware. That's the best news.
Compact code is going to start to matter more than it has. Of course,
"compact" will start to mean "First in a 32-bit address space" but hey
...

ron