If the Intel 80 core Tera-scale research chip actually works, why not just use a bunch of those?
It will require custom programming, but that should be easier than building (and maintaining) "dozens of racks and connected with an insane switch with up to 3456 ports."
I guess this will happen when EXOCHI is reality.
So, when we start hearing of exaFLOPS and zettaFLOPS ?

You know there is someone in the office sneaking in to play solitaire on that computer.

Why not go for the gusto and go with Yotta-scale? Then sit back and use the new tools to act as gods for all the little species..

In fact, making a super cluster does not represent that any company has the best technology; they have to be measured en terms of space, cost, and performance. Remember the old Virginia supercomputer center (Macs with PowerPC chips), it cost 5 million dollars and it was at the third place of top 500 of the world, but only cost 1/10 of the second place with only 10% of performance difference.. And even cost less that the 4th and 5th place.

In the first place, to build a supercomputer, a proporse is required. If we don't have a application in mind, why cluster thousands of intel 80 cores to catch yotta scales?

For sun, each core has up-to 8 threads.

and 20 years later (maybe 2025) we are going to have this power to play solitaire in 4D, surf internet2 and view Ultra-High-Hiper-Quality-Video (UHHQV :-) on our laptops!!, wow!!!

There's plenty of purpose and the need for high scale super computers. If super computers weren't available a lot more real world nuclear testing would occur here in the states or in some poor foreign country. Super computers allow scientist to model and study nuclear explosions and development for energy. There's also medical / gene related modeling that require intense computing. Oil companies, aircraft, and automotive industries even rent time to use these systems to determine strength and real world crashes and failures. Weather and earthquake forecasting is a major super computing task. These super computers have actually saved $100's of billions since many real world testing isn't required. Even the time saved is money saved. Many high-level simulation models can take hours, days, weeks, months, or even years without the processing power of super computers. When I was in college 16 years ago, it was stated that the DNA / RNA gene blue prints would take 20-30-50-100 years to decode. With the help of super computers complex gene therapy simulations can be developed in hours or days verus weeks, months, years.

Why do we build faster cars?For fun... do we need them to drive faster? The speed is limited... but... we build them..faster and faster.Now bugatti... in a few months a new mclaren F1 will come and will go to 410km/h, maybe in a few months it will reach 420 km/h and so on.

So the answer is: Let's see who has the biggest brain.we don't have a program to test it on full load...but ... we are building it.

I dont know about you, but using supercomputers to simulate nuclear bombs going out, has hardly any inpact for the great good for the nation?

I could be used to solve many other real world problems.

Lol, cause IBM does not want to use intel stuff, they have their own. moreover there is a huge energy reduction for blue gene, don't think intel will do so.

Custom Programming is the major cost, even the current HPC applications are not written in an efficient enough manner to utilize all of the processing speeds of BlueGene. Assembling and maintaining(physically) is trivial compared to the work that is required to get the operating system to not interfere with the jobs on supercomputers.
Also, the 80-core chip from Intel is single-precision floating point, just like all the graphics chips which can crank out 500GFlops - which isnt useful for many scientific apps other than rendering/visualization...
ExaScale systems are already on the drawing/discussion table and are slated for sometime in the next 10 years.

Super computers are also used to calculate real world disasters such as a tsunami or even such hurricanes as Katrina. Hospitals send off information for calculation done by Super computers. They can be used for calculating genomes for human diseases, you know like aids, and cancer. Lets not forget developing larger prime numbers for stronger encryption, or possibly cracking pre-existing encryptions for agencies such as the CIA. The limitless applications of the computing power can only be imagined. People asking "what is the point" or comparing this to building cars that "can go fast" are either dumb, or uneducated. Do you think IBM, SUN, or these other Universities think they are just making a multi million dollar computer for the hell of it? No they know they will have ample time to fest on the fruits of their super computing labor after they gain the initial recognition of having one of the top 500 computers.

Lets not forget that one of the main reasons why there is so much uncertainty about climate change is because the atmosphere and earth-atmosphere interactions are incredibly complicated systems. Anyone with a little knowledge of fluid dynamics should know this. We have not yet begun to exhaust computing power with this problem, so bring on the yottaflops.