In the recently published TOP500 ranking of supercomputers leader has changed again: Japanese car, which last year recognized the fastest, pushed supercomputer Sequoia, built by IBM.
Supercomputers - one of the most important indicators of technological armament. They allow you to predict the weather and stock prices, create new drugs and technologies, winning the war and to plan long-term development of the economy. Countries that have successfully put money into supercomputing, can more quickly than others to develop high technologies and significantly outperform those who do not have such machines.
The most powerful
U.S. continuously led among producing countries in the so-called supercomputers list TOP500 from 2004 until November 2010, when the system Cray Jaguar surprised many yield the palm car Tianhe-1A, created at the National Supercomputing Center in the Chinese city of Tianjin. Perhaps, for Americans especially sad that this unit was built largely on the basis of technology developed in the U.S. in particular, it uses GPUs at NVIDIA.

In June 2011, the Americans had already pushed into third place in world ranking Japanese car K Computer, designed with processors Fujitsu SPARC64 VIIIfx. The last time Japan could take the top of the list in 2004 - it was a supercomputer "Earth Simulator" firm NEC.
K Computer boasts more computing power than all five held the following stage machines together - despite the fact that the building at the time it was not even finished. According to the customer's K Computer, the Ministry of Education, Culture, Sports, Science and Technology of Japan, the designers were asked to aim to overcome the barrier of 10 petaflops (quadrillion floating point operations per second). This goal is reflected in the title: "C" - the first letter of the Japanese word "kei," which means 10 quadrillion (quadrillion - a million in the fourth degree, a unit with 24 zeros).
Second place left for the K Computer to 10.51 petaflops, and the third took the newcomer from the U.S. IBM BlueGene / Q system with a capacity of 8.15 petaflops.
The top ten were two Chinese cars: Tianhe-1A in fifth place, and Nebulae - the tenth.
The most powerful domestic supercomputer "Lomonosov", built at MSU by "T-Platforms," is in the new list in 22nd place with a capacity of just over 0.9 petaflops on the Linpack benchmark and the peak performance of about 1.7 petaflops. Unfortunately, this is the only Russian supercomputer, approaching twenty. Four other occupy 148, 182, 185 and 390 seats.
Interestingly division by country: United States - bessportny leader of the list, there are 253 developed a supercomputer out of five hundred. Asia's share - 121 vehicles in Europe - 107 systems. China represented 68 units, Japan - 34-axis, in the UK 25 supercomputers in France - 22, from Germany - 20, Russia - 5.
372 supercomputer (74.4%) are constructed on the basis of processor Intel, 63 systems (12.6%) - based chips AMD Opteron, 58 cars - processors, IBM Power. The 58 machines use graphics accelerators or co-processors: in 53 of them - chips NVIDIA, two - AMD Radeon, Cell in two and one - Intel MIC.
Chinese threat
The concern is, of course, not a performance, but the fact that this supercomputer built entirely of developed and manufactured in China processors instead of the usual chips Intel or AMD. Moreover, it was found that the size and energy efficiency of this machine is comparable to the twenty best supercomputers in the world, which indicates the highest level of engineering and technology.
Sunway BlueLight MPP built from 1600 processors ShenWei SW-3, each with sixteen cores, based on a 64-bit RISC-architecture, MIPS, clock frequency of 975 MHz. The official version is that the chips are designed based on micro Loongson / Godson, work on which is underway in China since 2001. According to others, it is a product of "reverse engineering" of the American chip DEC Alpha.
In any case, according to experts, these chips architecture for several generations behind contemporary American chips, but there is hardly any doubt that the Chinese government has been very generous to finance development in the near future to reduce the gap.
Exascale computing
While there is no generally accepted translation of the phrases in the Russian language, but its meaning is "to scale computing" ESCWA "ie quintillion (2 60 ). initiative aims Exascale computing, supported by the U.S. Department of Energy and its member National Nuclear Security Administration, is to increase the performance of supercomputers to thousands of times compared to the current level of 1-15 petaflops.
Some experts believe that a breakthrough can be expected by 2020, but so far there are insurmountable technical obstacles. The main one is that the higher the performance of the system, the greater its size, power, and, of course, the heat. Even modern supercomputers instantly turn into a heap of molten iron and plastic, if you suddenly refuse them a powerful cooling system that is more like a factory cold.
It is possible that a potential contender for the title of the most energy-efficient computer can be described above Sunway BlueLight MPP: According to official data, its power consumption is only one (1) MW. For comparison, the other "Chinese" Tianhe-1A, built on the basis of U.S. chip consumes is 4 MW, and the American Jaguar - more than five.
The productivity of Sunway is about 74% of the power Jaguar, which gives huge savings in operating costs, which reach a million dollars a year for each kilowatt of power only. Japanese K Computer consumes about 12.65 MW, which means almost 13 million dollars a year.
Meanwhile, the jump in scale, "ESCWA" 2020 is really necessary, if you consider the amount of data that we have to deal at this time. Judge for yourself: in 2004 the total amount of data stored online, estimated at about five exabytes. Already in 2009, increased by the same amount of data downloaded per month. Today, the monthly volume of about 21 exabytes. And that does not include data from other networks and sources - bank card networks, satellite tracking systems for weather and many others! Turn these enormous data streams into useful information can only supercomputers.
One way to dramatically improve the performance of computing is to use non-traditional supercomputer designs, components and materials. For example, at the Supercomputing Center of San Diego was built machine called Gordon using flash memory, to be exact, 1024 solid state drives Intel SSD-710, which can provide data to the order faster than conventional hard drives. In the list of TOP500 this the world's largest "flash" a 300 trillion bytes now occupies 62nd place with 0.28 petaflops performance on the test Linpack.
Spanish Supercomputing Center in Barcelona is going to use to build a supercomputer in 2014, "telephone" processors Tegra. The goal of the experiment - to reduce power consumption by 4-10% compared to the best energy-efficient systems.
You should not ignore the possibilities offered by virtual "cloud" supercomputer, which will be able to receive in rent to attend. Amazon has built a system in the autumn of 2011 and now it takes 72 th place in the TOP500. The network is Amazon so many located in different parts of the world data centers, that their combined power enough to provide "cloud" supercomputer multiple third-party services.
Anyway, the possession of any modern supercomputers to power. Therefore it is especially sad to see the list of five hundred most powerful supercomputers in the world just five Russian.
Supercomputers - one of the most important indicators of technological armament. They allow you to predict the weather and stock prices, create new drugs and technologies, winning the war and to plan long-term development of the economy. Countries that have successfully put money into supercomputing, can more quickly than others to develop high technologies and significantly outperform those who do not have such machines.
The most powerful
U.S. continuously led among producing countries in the so-called supercomputers list TOP500 from 2004 until November 2010, when the system Cray Jaguar surprised many yield the palm car Tianhe-1A, created at the National Supercomputing Center in the Chinese city of Tianjin. Perhaps, for Americans especially sad that this unit was built largely on the basis of technology developed in the U.S. in particular, it uses GPUs at NVIDIA.

As of June 2011, it was built only 672 of the planned 800 stands, with the system of 68,544 processors demonstrated in the Linpack benchmark performance of 8.162 petaflops in. The barrier was taken in November 2011: the car of 864 server racks with 88,000 processors shown in the same test performance in 10.510 petaflops, which is 93% of the theoretical peak power at 11 280 petaflops.
June 18, 2012 was presented to the next list of TOP500, the first place in which he again took a supercomputer from the U.S. - IBM Sequoia, installed at Lawrence Livermore National Laboratory, Lawrence Livermore, California. The machine shown in the Linpack benchmark performance in 16.32 petaflops.
Second place left for the K Computer to 10.51 petaflops, and the third took the newcomer from the U.S. IBM BlueGene / Q system with a capacity of 8.15 petaflops.
The top ten were two Chinese cars: Tianhe-1A in fifth place, and Nebulae - the tenth.
The most powerful domestic supercomputer "Lomonosov", built at MSU by "T-Platforms," is in the new list in 22nd place with a capacity of just over 0.9 petaflops on the Linpack benchmark and the peak performance of about 1.7 petaflops. Unfortunately, this is the only Russian supercomputer, approaching twenty. Four other occupy 148, 182, 185 and 390 seats.
Interestingly division by country: United States - bessportny leader of the list, there are 253 developed a supercomputer out of five hundred. Asia's share - 121 vehicles in Europe - 107 systems. China represented 68 units, Japan - 34-axis, in the UK 25 supercomputers in France - 22, from Germany - 20, Russia - 5.
372 supercomputer (74.4%) are constructed on the basis of processor Intel, 63 systems (12.6%) - based chips AMD Opteron, 58 cars - processors, IBM Power. The 58 machines use graphics accelerators or co-processors: in 53 of them - chips NVIDIA, two - AMD Radeon, Cell in two and one - Intel MIC.
Chinese threat
Americans today is not even a cautious concern suddenly burst through the 2010 first TOP500 Tianhe-1A, is actually made on technology from the U.S.. In October 2011, China officially introduced its new supercomputer Sunway BlueLight MPP, which is in the new list is not the highest position 26 with 7.95 petaflops on Linpack and a theoretical peak performance of 10.7 petaflops.
The concern is, of course, not a performance, but the fact that this supercomputer built entirely of developed and manufactured in China processors instead of the usual chips Intel or AMD. Moreover, it was found that the size and energy efficiency of this machine is comparable to the twenty best supercomputers in the world, which indicates the highest level of engineering and technology.
Sunway BlueLight MPP built from 1600 processors ShenWei SW-3, each with sixteen cores, based on a 64-bit RISC-architecture, MIPS, clock frequency of 975 MHz. The official version is that the chips are designed based on micro Loongson / Godson, work on which is underway in China since 2001. According to others, it is a product of "reverse engineering" of the American chip DEC Alpha.
In any case, according to experts, these chips architecture for several generations behind contemporary American chips, but there is hardly any doubt that the Chinese government has been very generous to finance development in the near future to reduce the gap.
Exascale computing
While there is no generally accepted translation of the phrases in the Russian language, but its meaning is "to scale computing" ESCWA "ie quintillion (2 60 ). initiative aims Exascale computing, supported by the U.S. Department of Energy and its member National Nuclear Security Administration, is to increase the performance of supercomputers to thousands of times compared to the current level of 1-15 petaflops.
Some experts believe that a breakthrough can be expected by 2020, but so far there are insurmountable technical obstacles. The main one is that the higher the performance of the system, the greater its size, power, and, of course, the heat. Even modern supercomputers instantly turn into a heap of molten iron and plastic, if you suddenly refuse them a powerful cooling system that is more like a factory cold.
It is possible that a potential contender for the title of the most energy-efficient computer can be described above Sunway BlueLight MPP: According to official data, its power consumption is only one (1) MW. For comparison, the other "Chinese" Tianhe-1A, built on the basis of U.S. chip consumes is 4 MW, and the American Jaguar - more than five.
The productivity of Sunway is about 74% of the power Jaguar, which gives huge savings in operating costs, which reach a million dollars a year for each kilowatt of power only. Japanese K Computer consumes about 12.65 MW, which means almost 13 million dollars a year.
Meanwhile, the jump in scale, "ESCWA" 2020 is really necessary, if you consider the amount of data that we have to deal at this time. Judge for yourself: in 2004 the total amount of data stored online, estimated at about five exabytes. Already in 2009, increased by the same amount of data downloaded per month. Today, the monthly volume of about 21 exabytes. And that does not include data from other networks and sources - bank card networks, satellite tracking systems for weather and many others! Turn these enormous data streams into useful information can only supercomputers.
One way to dramatically improve the performance of computing is to use non-traditional supercomputer designs, components and materials. For example, at the Supercomputing Center of San Diego was built machine called Gordon using flash memory, to be exact, 1024 solid state drives Intel SSD-710, which can provide data to the order faster than conventional hard drives. In the list of TOP500 this the world's largest "flash" a 300 trillion bytes now occupies 62nd place with 0.28 petaflops performance on the test Linpack.
Spanish Supercomputing Center in Barcelona is going to use to build a supercomputer in 2014, "telephone" processors Tegra. The goal of the experiment - to reduce power consumption by 4-10% compared to the best energy-efficient systems.
You should not ignore the possibilities offered by virtual "cloud" supercomputer, which will be able to receive in rent to attend. Amazon has built a system in the autumn of 2011 and now it takes 72 th place in the TOP500. The network is Amazon so many located in different parts of the world data centers, that their combined power enough to provide "cloud" supercomputer multiple third-party services.
Anyway, the possession of any modern supercomputers to power. Therefore it is especially sad to see the list of five hundred most powerful supercomputers in the world just five Russian.






No comments:
Post a Comment