Tuesday, July 26, 2005

Simulator 2

It's in the news now, so I think I can safely blog about this without fear of causing embarrassment:

Here is today's Japan Times article on plans for a new 10PFlop supercomputer, Simulator 2. Yes, that's 10 petaflops, a whopping 250 times faster than the Earth Simulator and up to 73 times faster than the current fastest Blue Gene system. Plans are a bit vague, but with a planned switch-on date of 2010 it would probably leapfrog all the competition by quite a way, as the Earth Simulator did back in 2001.

You can find out more about it here and here, but only in Japanese!

I believe the research remit for ths use of this new machine might be slightly broader than the current Earth System Science focus (hence Simulator 2, not Earth Simulator 2). The plans still seem pretty vague but my experience here suggests they don't usually publicise things that are likely to fall through. Whether or not I'll still be here to use it is very much less certain...


Anonymous said...

I am no more sure than you are about this issue, but, unfortunately (from the vocational viewpoint of climate scientists), climate simulation may be no focus of the coming simulator but just one of hundreds of competing subjects. The document 001.pdf you mentioned explains application in molecular dynamics for material engineering, and 002.pdf mainly deals with nuclear plant engineering and only occasionally (perhaps less than 5 %) ocean circulation modelling.

In the meeting on Friday 19th (you were there too), a presenter communicated a story by his colleague in the Japan Meteorological Agency. JMA submitted an informal proposal to use the new simulator (probably for weather application rather than climate), but MEXT turned it down. It seems that it was precisely because the proposed usage conforms the architecture of the current Earth Simulator (cluster of vector computers), just scaled up. The simulator project is essentially the national project for enhancing computer technology, and it must have an innovative architecture.
The innovative architecture here means synergy (I do not know no more) between vector and scalar architectures. It may be beneficial for ecosystem simulations, for example. But promoters of cloud-resolving global models around us consider that scale separation is inappropriate for the atmosphere. It is difficult to conform with the desire of MEXT.

A. Sumi, professor at Univ. Tokyo and the leader of the global warming projection project which we collaborate, said that it may not be worthwhile making negotiation to participate in the national project. Rather, he said, we should try to get fund to purchace computer power which will become available commercially.

Anonymous said...

Excuse me, this is a correction to a trivial mistake in my previous commment: The meeting was on Tuesday, 19th of July, not Friday.

James Annan said...


Thanks for the comment and translation - as you guessed, I didn't get all of it :-)

Whether or not "we" get to use S2, it still seems a generally positive sign. After all, a 1% share of a 10PFlop machine is no better than 100% of a dedicated 100TFlop system. With the obvious international and political importance of climate research, I am confident this area of research will not be left to shrivel and die. The existing ES will still be useful for several years to come, of course.