[adorno]

impressive 16.32 petaflop/s on the Linpack benchmark using 1,572,864 cores

Impressive would be to have the 16.32 petaflops achieved by a one CPU single core system.

In a few weeks, the IBM system will be eclipsed by some other system, where somebody put together a system with just a few more cores. That's the super-computing paradigm nowadays, where a supercomputer is defined by the number of cores that someone can string together.

[HeartlandOfAmerica]

In a few weeks, the IBM system will be eclipsed by some other system,

lol. yeah. SOme other system using American technology. Did you see the listing? IBM, Intel, SPARC64, HP, NVIDIA, ATI.

[zeugma]

Impressive would be to have the 16.32 petaflops achieved by a one CPU single core system.

You're right, that would be pretty impressive.

In a few weeks, the IBM system will be eclipsed by some other system, where somebody put together a system with just a few more cores. That's the super-computing paradigm nowadays, where a supercomputer is defined by the number of cores that someone can string together.

You know that there is a heck of a lot more engineering that goes into these systems that just plugging a bunch of processors into a bank of motherboards don't you? One of the reasons IBM does so well in this type of thing is that they have a lot of experience with how to move massive amounts of data reliably across a high-speed backplane. This is something IBM learned how to do building mainframes for the last many decades.

This type of high-end speed test doesn't have much to do with us in our everyday lives, but there are problems out there that will eat all the horsepower they can be given. You create the necessary computational horsepower any way you can.