Slashdot Mirror
Sandia Wants To Build Exaflop Computer
Dan100 brings us an announcement that Sandia and Oak Ridge National Laboratories are setting their sights on an exaflop supercomputer. Researchers from the two laboratories jointly launched the Institute for Advanced Architectures to facilitate development. One of the problems they hope to solve is how to provide each core of each processor with enough data so that cycles aren't going to waste.
"The idea behind the institute — under consideration for a year and a half prior to its opening — is 'to close critical gaps between theoretical peak performance and actual performance on current supercomputers,' says Sandia project lead Sudip Dosanjh. 'We believe this can be done by developing novel and innovative computer architectures.' The institute is funded in FY08 by congressional mandate at $7.4 million."
You don't usually run one program on these type of systems. The compute cycles are bidded out to researchers and they get x number of compute hours. The system is partitioned out to a few nodes and given to the researcher to run their codes on. You could have on a system like this hundreds of jobs running simultaneously. Also, with the tens of thousands of cores needed to reach this status, a node failure, or other hardware failure is inevitable. Right now if a node fails in the middle of the job, everything is lost from the last checkpoint. The chances of failures impeding work go up greatly the more nodes and cores you run the job on.
7 million isn't alot for a datacenter with a supercomputer housing a novel architecture. They'll need infiniband or gig fiber, and other high end equipment. That in itself will take 2 million at least. I dunno. Maybe they'll do a low-rent google and call it unique.
Try Climate & Weather Codes, Fusion, Combustion, CFD, Bio (genomics), and
a host of large science/engineering, partial differential equation=based applications
requiring the solution of large systems of matrix equations,... Check out:
http://www.ornl.gov/info/press_releases/get_press_release.cfm?ReleaseNumber=mr20061025-00
What's past is NOT ALWAYS prologue for the future!
We have elected a republican president... twice. In fact it is this republican president that has put the emphasis on American primacy in the super computing arena. Something called the earth simulator out of Japan put us in our place. This president opened up funding and in effect mandated the classes of systems you see today being built at NASA, DoE facilities, and academia. But, go on with your blind hatred and closed mindedness.
Reading the article, the goal is nowhere near building a real exaflop computer, but more about thinking about issues (like processor data feeding).
In a year and a half, we shouln't have more than 100 GFlops per socket, which means that you will still need 10 millions of processors (not cores!) to achieve the exaflop computer. No chance to build a cluster that big (at least these years).
The all-times progression of the top500 shows that exaflop computers should arrive around year 2020, definetly not tomorrow. (x10 every ~4 years, 2008:1 PF, 2012:10 PF, 2016:100 PF, 2020:1 EF)
Many people have worked on the area which they think AI is lacking in, but until we understand the brain (or even come up with a good definition of "intelligence") I don't think we'll get very far. But who knows, keep on looking!
There are a lot of uses for extra computing power though, it's not like we've reached a point where we have too much. Protein folding and climate models are the first that come to mind, but I'm sure there are many others. Companies aren't building these things for fun.
// MD_Update(&m,buf,j);
If your work tends to be I/O bound... it doesn't belong on an exaflop cluster. You know BlueGene/L? Most of its nodes don't even talk directly to the storage system--they're connected to special I/O nodes which then talk to the storage system.
Scientific computing doesn't really deal with THAT much data. The scientists here at Sandia (yeah I work at Sandia CA) think they are just HUGE data creators. "We generate a PETABYTE per YEAR!" they say... not realizing that a petabyte is a drop in the bucket for the guys running these systems. As a colleague from LLNL said the other day, a petabyte isn't even worth charging for--they've got that much storage available in the tapes lying around the machine room.
Klingon programs don't timeshare, they battle for supremacy.