'K' Computer Now Fastest In The World

[Sean Evans]

Japanese 'K' Computer Now Fastest In The World

The world's most powerful computer is now in Japan, called K Computer

According to The New York Times, the computer, called "K," is three times faster than the next fastest computer, a Chinese computer, Tianhe-1A, that previously held the highest spot.

K, built by Fujitsu, can make 8.2 quadrillion calculations each second. "K" stands for "Kei," meaning 10 quadrillion, or the number of calculations per second researchers hope K will one day be able to perform per second. K's abilities are the equivalent of linking one million desktop computers.

K is made out of 672 computer racks, with 68,544 computer processing units (CPUs) but the lab plans to raise that number to 800 racks, a move that will make the machine even speedier and more powerful.

Scientists expect that the computer will be used in fields including climate research and disaster prevention, as well as medicine. The machine is scheduled for deployment in 2012.

"Bringing together hundreds of thousands of components to quickly launch such a massive-scale computing system-which would have been nearly impossible using conventional technologies-requires an incredible level of reliability," wrote Michiyoshi Mazuka, Chairman of Fujitsu Limited in a press release. "I believe that this reliability is truly the pinnacle of Japanese manufacturing."

The U.S. has five of the world's ten most powerful computers, with a computer in Oak Ridge, Tennessee coming in third on the overall list.

[S.Taylor]

Obvious question, for those who don't know already, including me....

Would this change the whole game of chess, or atleast 500 elo above the top, like say, Houdini on fastest proccessor. Or, Rybka on Cluster?

[S.Taylor]

I'm not up to date with my facts and figures, but i would have expected that, numbering in the quadrillions per second, would make any cluster like about as fast as a snail vs a plane.

[georgerifkin]

I think I've read somewhere in the past that those supercomputers are almost useless for chess programs because of the comunication delays among the computers linked or something like that

[bob]

Obvious question, for those who don't know already, including me....

Would this change the whole game of chess, or atleast 500 elo above the top, like say, Houdini on fastest proccessor. Or, Rybka on Cluster?

It is a giant cluster. 'nuff said...

[FlavusSnow]

Something to note: these supercomputers are ranked by their floating point benchmarks. Its a little bit like measuring a GPU vs a CPU. I understand that these huge clusters are often fully capable running integer calcs, but certainly optimized for FLOPS.

Its my experience that chess performance is more closely related to the integer performance of the CPU. I'm not quite sure I understand why the ALU matters more than anything else, but experience tells me it does.

All that to say that there is no way to know how something like this would perform at chess, and it'd certainly take some custom programming function on the cluster or NUMA architecture of these behemoths. Let alone to get an engine to search efficiently on such a machine. I'd think it would take new algorithms, or at least applying current algorithms in new ways.

I could theorize on what would work better, but people here don't care about anything except what presently works the best on today's household hardware.

[fern]

I wonder if with such brute force any program, even a 1960 clunker, could play at master level...
Perhaps only could be enough a table of material values and some mininal prunni8ng.

Or not? .

Bob?

Fern