on august 10 2011, news came out that IBM drops 10Flop NCSA Supercomputer project because of technical complexity and cost. Now they're going for something 10 times faster. Big Blue going after Big Dream.
"The patent claims that each processing node will consume about 30 watts of power, which puts the 107 PFlop/s system at only 15.7 MW. That is rather impressive for a system with more than 8 million CPU cores."
That's extremely impressive given the level of peak performance. Many current supercomputers in the 1 Pflop range can consume ~5-10 MW.
100 PFLOP/s sounds impressive (although I don't know much about what PFLOP/s exactly refers to) and that is a *** ton of cores..but I'd be more impressed to see that much processing power with way fewer cores. Anyone can add another processor and circuit board to the rack, but improving technology and packing more power into one CPU/board is what its all about. right?
Damn.... and I though that the i5 2500K was already fast enough... (from what I've seen) This REALLY makes my 1st gen quad i5 750 look shitty, (and 8GB DDR3 1600) even when I upgrade the gpu to a ati 5770 or something...
[citation][nom]livebriand[/nom]Damn.... and I though that the i5 2500K was already fast enough... (from what I've seen) This REALLY makes my 1st gen quad i5 750 look shitty, (and 8GB DDR3 1600) even when I upgrade the gpu to a ati 5770 or something...[/citation]
Yes, some super computers (more as time goes on) do use x86 processors, but the processors used in this planned and current IBM supercomputers use the Power architecture. Keep in mind that the Power architecture is a completely difference beast than x86, designed with with a much different set of goals in mind.
Why can't ordinary software developers take a page or two off these highly parallel supercomputers? If the software that runs on these things can take advantage of thousands or millions of cores at once, why can't, for example, game developers even take advantage of four cores properly?