K computer which cost $1.2 billion has been overtaken by the $97 million Titan. If this doesn't exemplify all that's wrong with Japan Inc. I dunno what does! Imagine yourself building a gaming reg only for it to be humiliated 18 months later by a pocket calculator.
Technology does improve and it is not fair to compare a 2011 supercomputer with a 2012 supercomputer, but technology never improved at the rate of twice as fast for one tenth the price in one year, that would be exactly like having a $1200 computer beaten a year later by a $97 calculator in everything. Titan is not only twice as fast but also more than twice as energy efficient. A better comparison to K might be the IBM Sequoia, both are Massively Parallel Processor (MPP) designs and designed by one company using its own hardware from the CPU all the way to the interconnect as opposed to most supercomputers that use off the shelf components and are designed as clusters. Sequoia cost about half as much as K, doesn't use any accelerators unlike Titan and has similar yet not equivalent design to the K.
Then there is this, the K is very expensive to run because it is energy inefficient. The K was designed for another time and age apparently, nuclear era Japan when electricity was cheap and plentiful, no one imagined how things can change. This wouldn't be worrying if it weren't for the strong push towards energy efficiency that everyone jumped on except Fujitsu.
K has a relatively low performance/watt and low performance/cost. The total cost of K is estimated at $1.2 billion with a yearly running cost estimated at $10 million. K ranks 85th on Green 500 list with 830.18 MFLOPS/W. Titan ranks 3rd on the Green 500 at 2,142.77 MFLOPS/W. Titan not only achieves high performance/watt but also high performance/dollar. At $100 million the cost of Titan is a fraction of the cost K, Titan's annual running costs are $9 million. Fujitsu recent announcement of adopting Xeon Phi for its future computers show that K might be the last of the homogenous all-CPU supercomputers and SPARC64 is not long for this world, not if Oracle gets its way.
Technology does improve and it is not fair to compare a 2011 supercomputer with a 2012 supercomputer, but technology never improved at the rate of twice as fast for one tenth the price in one year, that would be exactly like having a $1200 computer beaten a year later by a $97 calculator in everything. Titan is not only twice as fast but also more than twice as energy efficient. A better comparison to K might be the IBM Sequoia, both are Massively Parallel Processor (MPP) designs and designed by one company using its own hardware from the CPU all the way to the interconnect as opposed to most supercomputers that use off the shelf components and are designed as clusters. Sequoia cost about half as much as K, doesn't use any accelerators unlike Titan and has similar yet not equivalent design to the K.
Then there is this, the K is very expensive to run because it is energy inefficient. The K was designed for another time and age apparently, nuclear era Japan when electricity was cheap and plentiful, no one imagined how things can change. This wouldn't be worrying if it weren't for the strong push towards energy efficiency that everyone jumped on except Fujitsu.
K has a relatively low performance/watt and low performance/cost. The total cost of K is estimated at $1.2 billion with a yearly running cost estimated at $10 million. K ranks 85th on Green 500 list with 830.18 MFLOPS/W. Titan ranks 3rd on the Green 500 at 2,142.77 MFLOPS/W. Titan not only achieves high performance/watt but also high performance/dollar. At $100 million the cost of Titan is a fraction of the cost K, Titan's annual running costs are $9 million. Fujitsu recent announcement of adopting Xeon Phi for its future computers show that K might be the last of the homogenous all-CPU supercomputers and SPARC64 is not long for this world, not if Oracle gets its way.
