Slashdot Mirror
Cell-based "Roadrunner" Tops Elusive Petaflop Mark
prunedude writes "The NY times is reporting that an American military supercomputer, assembled from components originally designed for video game machines, is more than twice as fast as the previous fastest supercomputer, the I.B.M. BlueGene/L. To put the performance of the machine in perspective, Thomas P. D'Agostino, the administrator of the National Nuclear Security Administration, said that if all six billion people on earth used hand calculators and performed calculations 24 hours a day and seven days a week, it would take them 46 years to do what the Roadrunner can in one day."
As a software developer who's worked on the Lab's previous ASC machines (Blue Mountain, Q, Lightning) I can say that once the calculation is run to get a machine atop Jack Dongarra's gee-golly list, it's partitioned, segmented, divided, and subjected to such crappy resource management that if I could trade the entire machine for a pair of coupled 8-core Mac Pros I'd do it in a heartbeat.
The real PITA with these machines is that the powers that be are trying to kill two birds with one stone: they want an R&D platform for advanced computing, but they also want to certify an aging and untestable nuclear stockpile. That rather requires a fairly static platform, and so far our experience with ASC has been that when a machine hits that sweet state, they yank it and give us the next one.
Learn to spell: nickel, missile, lose, solely, amendment, speech, kernel, probably, ridiculous, deity, hierarchy, versus
I suspect the first example of this happening was trying to estimate how many angels could fit on the head of a pin.
Other meaningless analogies could be:
The simple fact is that a petaflop computer works faster than humans can conceive and any kind of analogy cannot be comprehended.
myke
Mimetics Inc. Twitter
From what I've heard (I'm no expert), these processors are good at certain types of calculations, but horrible at others. Ask a Cell to run Folding@Home, and it'll be blazing. Ask it to do general-purpose computing, though, and you'll quickly see the strength of other processors.
The biggest problem is not to acquire the hardware (the graphics boards can be bought easily enough and some manufacturers make motherboards that will host 4 of them), the problem is in the software.
As you probably know your current desktop computer executes software 'in parallel' on multiple cores. Unless that software is written to use multiple cores then the parallelization will probably take place at the task level, in other words, unless special care was taken during the development each core will be executing a different process (or parts of a process, slicing itself up between multiple processes as long as the number of tasks is greater than the number of cpus).
The cell processor and the graphics cards you are talking about (I assume you refer to the nvidia 'cuda' architecture) are not your ordinary processor (though the cell comes a lot closer). They need special software to get the maximum out of their parallel features, which means that you can only exploit that speed when you are trying to solve a particular class of problems.
The issue is that not every problem is easily adapted to a parallel solution, and for some classes of problems it's simply impossible. For general purpose computing a general purpose cpu (with or without multiple cores) is good enough.
If you have a problem that you can solve in a parallel way (say computing chess moves, analyzing genetic strings, ray tracing an image or fluid dynamics, add your own examples here) then it makes sense to invest the extra time to code the problem up in such a way that if you have lots of cpus that they can each work on a small part of the problem.
The graphics cards are closer to a traditional vector processor than the cell, and thus even less suitable to be used as general purpose cpus.
MP3 Search Engine