TOP500. The org organization today announced the 36th ranking of the world's top 500 supercomputers. China’s Tianhe No. 1 A, which appeared only at the end of last month, took the top spot with no suspense, and this is the first time in Chinese history that this ranking has been held. On top of the top spot.
The "Tianhe 1" before the improvement has been ranked fifth and seventh consecutively. The newly upgraded Tianhe No. 1A is based on the NUDT YH Cluster. The hardware is equipped with an Intel Xeon X5670 2.93GHz six-core processor (32nm Westmere- EP), China's self-developed Feitun FT-1000 eight-core processor, NVIDIA Tesla M2050 high-performance computing card, 224TB memory, proprietary interconnect architecture, Linux operating system, a total of 186,368 cores, Linpack maximum performance 2.566PFlops (per second Petaflops), peak performance 4.701 PFlops, system efficiency 54.6%.
Tianhe No. 1 A is located in the National Supercomputing Center in Tianjin. It has been fully operational immediately after its completion. It is mainly used to implement large-scale scientific calculations and is also an open access system.
For the past two consecutive championships, the Jaguar created by Cray Corporation for the computational science center of the Oak Ridge National Laboratory in the United States has remained unchanged in terms of hardware configuration and computing performance over the past year and has now slipped to second place.
Nebulae, China's other supercomputer that had won the runner-up position half a year ago, did not make the upgrade. Now it ranks third, but for the first time we know its total power consumption: 2,580 kilowatts. In fact, the Tianhe No. 1 A's power consumption is only 4040 kilowatts, which is 42% lower than the Jatropha's 69,950 kilowatts. It is already very energy-efficient.
In addition to Tianhe No. 1A and Nebula, there are also three sets of domestic supercomputers that were selected before the final 100, namely:
- 28th "Mole-8.5", Institute of Process Engineering, Chinese Academy of Sciences / Taian Mole-8.5 Cluster, Intel Xeon L5520 2.26GHz Quad-Core Processor, 33120 Cores, NVIDIA Tesla Computing Card, Maximum Performance 207.3TFlops, Peak Performance 1183.44TFlops ;
- The 35th "Magic Cube", Dawning 5000A system, AMD Opteron 1.9GHz quad-core processor, 30720 cores, 120TB memory, Windows HPC 2008 operating system, maximum performance 180.6TFlops, peak performance 233.472TFlops;
- The 68th "DeepComp 7000", Lenovo HS21/x3950 cluster, Intel Xeon 2.93/3.00GHz quad-core processor, 12,216 cores, maximum performance 102.8TFlops, peak performance 145.965TFlops.
It is quite encouraging that most of the five most powerful domestic supercomputers in the front are from domestic companies such as Shuguang and Lenovo and scientific research institutions such as the Chinese Academy of Sciences, while most of the 36 sets are IBM systems, and a few are from HP. Inspur has fully proved China's independent development capability of high-performance computing.
In this Top 500 ranking, China has a total of as many as 41 systems shortlisted, the share rose to 8.20%, thrown off the British, French, German and other strong and second only to the United States, ranked second in the world, and six There were still 24 sets and 4.80% before the month. Although the United States is still far ahead, but the total number of systems from 282 to 275 sets again, the share of 55.00%. In addition, Hong Kong, China continues to be a set.
It is worth mentioning that the “Tsubame 2.0†of Tokyo Institute of Technology in Japan also entered the TOP10 historically and reached the fourth place. It also uses the Intel Xeon X5670 six-core processor and NVIDIA Tesla computing card, with a maximum performance of 1.192 PFlops. It also broke the 1PFlops mark. After two and a half years of development, there have been seven sets of supercomputers that cross this threshold, including three sets in the United States, two sets in China, one set in Japan and one in France.
From the front we have also seen that GPU acceleration has become a major trend in high-performance computing. Including China's Tianhe No. 1A, Nebula, and Japan's Tsubame 2.0, a total of 17 systems in the Fortune 500 use GPU acceleration, including 10 sets of NVIDIA Tesla GPU chips, one set of AMD Radeon GPU chips, and six sets of IBM. Cell chip.
Some other points:
- In terms of vendor share, Cray, which has a long history of supercomputing, gained another two percentage points, surpassing HP, but still far from IBM. Cray's XT and XE systems are very popular and powerful, occupying four positions in the top ten.
- HP fell to third place in terms of volume, but overall performance was still higher than Cray and ranked second.
-The Intel processor share was as high as 79.6%, which appeared in 398 systems, but a decrease of 1.6 percentage points (8 systems) from six months ago. AMD’s share has increased by two percentage points, currently at 11.4%. IBM lost two sets and dropped to 8.0%.
- The penetration rate of quad-core processors has reached 73%, and there are also 19% with six cores or even more.
- The top 500 entries have risen from 24.7 TFlops to 31.1 TFlops, and the 500th is currently ranked at 305 half a year ago.
- The total performance of the top 500 increased from 27.6 PFlops a year ago to 34.2 PFlops six months ago to 44.2 PFlops.
- In terms of power consumption, only 25 systems have been confirmed to exceed 1,000 kilowatts. IBM's newly created BlueGene/Q system (prototype) created a new energy efficiency record of 1680 MFlops/W, more than twice the number two.
- The 500 average power consumption is 447 kilowatts, and the average energy efficiency is 195 MFlops/W, compared to 150 MFlops/W six months ago.
- The average power consumption of the top ten rises from 2.89 megawatts to 3.2 megawatts, and the average energy efficiency is reduced from 300 MFlops/W to 268 MFlops/W.
The "Tianhe 1" before the improvement has been ranked fifth and seventh consecutively. The newly upgraded Tianhe No. 1A is based on the NUDT YH Cluster. The hardware is equipped with an Intel Xeon X5670 2.93GHz six-core processor (32nm Westmere- EP), China's self-developed Feitun FT-1000 eight-core processor, NVIDIA Tesla M2050 high-performance computing card, 224TB memory, proprietary interconnect architecture, Linux operating system, a total of 186,368 cores, Linpack maximum performance 2.566PFlops (per second Petaflops), peak performance 4.701 PFlops, system efficiency 54.6%.
Tianhe No. 1 A is located in the National Supercomputing Center in Tianjin. It has been fully operational immediately after its completion. It is mainly used to implement large-scale scientific calculations and is also an open access system.
For the past two consecutive championships, the Jaguar created by Cray Corporation for the computational science center of the Oak Ridge National Laboratory in the United States has remained unchanged in terms of hardware configuration and computing performance over the past year and has now slipped to second place.
Nebulae, China's other supercomputer that had won the runner-up position half a year ago, did not make the upgrade. Now it ranks third, but for the first time we know its total power consumption: 2,580 kilowatts. In fact, the Tianhe No. 1 A's power consumption is only 4040 kilowatts, which is 42% lower than the Jatropha's 69,950 kilowatts. It is already very energy-efficient.
In addition to Tianhe No. 1A and Nebula, there are also three sets of domestic supercomputers that were selected before the final 100, namely:
- 28th "Mole-8.5", Institute of Process Engineering, Chinese Academy of Sciences / Taian Mole-8.5 Cluster, Intel Xeon L5520 2.26GHz Quad-Core Processor, 33120 Cores, NVIDIA Tesla Computing Card, Maximum Performance 207.3TFlops, Peak Performance 1183.44TFlops ;
- The 35th "Magic Cube", Dawning 5000A system, AMD Opteron 1.9GHz quad-core processor, 30720 cores, 120TB memory, Windows HPC 2008 operating system, maximum performance 180.6TFlops, peak performance 233.472TFlops;
- The 68th "DeepComp 7000", Lenovo HS21/x3950 cluster, Intel Xeon 2.93/3.00GHz quad-core processor, 12,216 cores, maximum performance 102.8TFlops, peak performance 145.965TFlops.
It is quite encouraging that most of the five most powerful domestic supercomputers in the front are from domestic companies such as Shuguang and Lenovo and scientific research institutions such as the Chinese Academy of Sciences, while most of the 36 sets are IBM systems, and a few are from HP. Inspur has fully proved China's independent development capability of high-performance computing.
In this Top 500 ranking, China has a total of as many as 41 systems shortlisted, the share rose to 8.20%, thrown off the British, French, German and other strong and second only to the United States, ranked second in the world, and six There were still 24 sets and 4.80% before the month. Although the United States is still far ahead, but the total number of systems from 282 to 275 sets again, the share of 55.00%. In addition, Hong Kong, China continues to be a set.
It is worth mentioning that the “Tsubame 2.0†of Tokyo Institute of Technology in Japan also entered the TOP10 historically and reached the fourth place. It also uses the Intel Xeon X5670 six-core processor and NVIDIA Tesla computing card, with a maximum performance of 1.192 PFlops. It also broke the 1PFlops mark. After two and a half years of development, there have been seven sets of supercomputers that cross this threshold, including three sets in the United States, two sets in China, one set in Japan and one in France.
From the front we have also seen that GPU acceleration has become a major trend in high-performance computing. Including China's Tianhe No. 1A, Nebula, and Japan's Tsubame 2.0, a total of 17 systems in the Fortune 500 use GPU acceleration, including 10 sets of NVIDIA Tesla GPU chips, one set of AMD Radeon GPU chips, and six sets of IBM. Cell chip.
Some other points:
- In terms of vendor share, Cray, which has a long history of supercomputing, gained another two percentage points, surpassing HP, but still far from IBM. Cray's XT and XE systems are very popular and powerful, occupying four positions in the top ten.
- HP fell to third place in terms of volume, but overall performance was still higher than Cray and ranked second.
-The Intel processor share was as high as 79.6%, which appeared in 398 systems, but a decrease of 1.6 percentage points (8 systems) from six months ago. AMD’s share has increased by two percentage points, currently at 11.4%. IBM lost two sets and dropped to 8.0%.
- The penetration rate of quad-core processors has reached 73%, and there are also 19% with six cores or even more.
- The top 500 entries have risen from 24.7 TFlops to 31.1 TFlops, and the 500th is currently ranked at 305 half a year ago.
- The total performance of the top 500 increased from 27.6 PFlops a year ago to 34.2 PFlops six months ago to 44.2 PFlops.
- In terms of power consumption, only 25 systems have been confirmed to exceed 1,000 kilowatts. IBM's newly created BlueGene/Q system (prototype) created a new energy efficiency record of 1680 MFlops/W, more than twice the number two.
- The 500 average power consumption is 447 kilowatts, and the average energy efficiency is 195 MFlops/W, compared to 150 MFlops/W six months ago.
- The average power consumption of the top ten rises from 2.89 megawatts to 3.2 megawatts, and the average energy efficiency is reduced from 300 MFlops/W to 268 MFlops/W.
Dongyuan Syscooling Technology Co., Ltd. , http://www.syscooling.com