Supercomputers

[THCi]

With the current fast development of PC/Desktop/Consumer level machines, why is it we dont get the same fast paced development of supercomputers?

One facter would be, of course, expence. But, still that doesnt rule them out for a small number of firms.

What are your views on the subject?

[Agent]

I think there are super computers out there that we havnt even heard of, or are likely to anytime soon. I doubt they are going to release much to the public.

[joshwa]

what kind of super computers?

like the 10,000 CPU Opteron Cray system ?

http://www.theinquirer.net/?article=10668

[THCi]

Yeah, that one inspired this debate.

And, of course htere are supercomputers that we dont know about, its simple: Gov't organisations.

There are gonna be supercomputers that the organisations that own them would rather destroy them than tell the public about them.

But, that isnt what this thread is about.

The supercomputers that WE do know about dont seem to be made/built/designed as quickly as thier desktop counter parts. why?

[Zathras]

It's simple economics really. The consumer/business market is huge therefore a lot of development goes on for normal cpus like the P4 and Athlon families. Supercomputers no longer use specialised CPUs but rather distributed clusters of 'normal' CPUs due to improvements in inter-cluster networking and job allocation software and the relative costs. A large supercomputer requires a lot of planning, testing, design etc so it's no wonder they don't get rolled out as quickly and when they do they don't always use a cluster of the fastest available chips.

[Agent]

Maybe its the Supply and demand chain kicking in ?
There is a high demand for desktop CPU's, so a lot of R&D would go into them, with super computers, its going to be a very small demand, so much less R&D would be put in and progress slower.
I think that could be one of the main reasons.

[Rys]

I think the apparent slowness of development in supercomputers is perceived, rather than real.

Depends what you mean by development. It's fast paced by nature on the desktop home PC market, but in terms of development to take the next step in supercomputing terms, I think the efforts are just as big, just not as visible.

Plus the fact it takes time to properly commision, spec, build, test, retest, pay for and deploy a large supercomputer. In the order of years.

Two different world, incomparable ones at that if you ask me. Try finding an equivalent to Moore's law in the supercomputer space, I bet it exists, but is just in a different part of the system, other than the CPU. Something like terraflop increase per year, or some other metric.

Rys