Virginia Tech may be souping up System X with special-order 2.3 GHz Xserves, and the U.S. Army and COLSA may be hitching up even more Xserves than Virginia Tech, but so far Mac-based supercomputers are a distinct minority in the field, and none of the Mac clusters even comes close to the current Big Daddy of the teraflop scene: Japan's $350 million Earth Simulator, which has held the title of "World's Fastest Supercomputer" for two full years, now-- practically forever in the fast-paced world of LINPACK benchmarks. Think of it this way: when System X was the third-fastest supercomputer on the planet, it was cranking out 10.28 teraflops of raw, crunchy power; the Earth Simulator, by comparison, spits out 35.86. Two words: Yee and ikes.

But while there isn't yet a Mac-based challenger to the Earth Simulator's superiority, faithful viewer RevMark informs us that there's something almost as good: IBM's BlueGene/L, a cooperative supercomputing project designed to scale to, frankly, totally ridiculous levels. In the current TOP500 list, BlueGene/L prototypes occupy the fourth and eighth spots, scoring 11.68 and 8.66 teraflops respectively-- not too shabby. But remember, those are just prototypes, using far fewer processors than IBM's final systems will employ-- and according to Reuters, a BlueGene/L system has now benched out at a face-melting 36.01 teraflops, which is just enough to steal the crown from Japan's Finest.

Why is this potentially good for Apple? Well, because while BlueGene/L isn't a Mac, its processors are PowerPCs, albeit ones that aren't quite the same as the G5s that IBM squeezes out for Apple. Still, the world's fastest supercomputer running PowerPC chips could well lend Macs more street cred for being the blazing speed demons they are, and slowly persuade the public that "Intel Inside" isn't all it's cracked up to be. If nothing else, science geeks (who are already being drawn to the Mac in ever-increasing numbers because of price/performance ratios and the UNIXy goodness of Mac OS X) ought to come a-runnin' with their grant checks ready to endorse.

Oh, and by the way, the 36.01 teraflop BlueGene/L unit isn't the full system, either. Remember that fourth-ranked prototype unit that scored 11.68 teraflops? Well, that represents one sixteenth of the primary BlueGene/L being built for Lawrence Livermore National Laboratory. What's more, that prototype uses 8,192 custom 500 MHz PowerPC 440s, which is an older chip; the eighth-ranked system used half as many processors running at 700 MHz to score 8.66 teraflops. Do a little naïve math, factor in the possibility of faster processors, and you find that the Livermore cluster ought to score at least 300 teraflops when it's done next year; IBM says it'll reach 360. Well, if you're going to outscore the world's fastest, why not do it by a factor of ten?