"Tianhe-2" is a supercomputer system developed by National University of Defense Technology, ranking first with its excellent performance of peak computing speed of 549 million times per second and continuous computing speed of 339 million times per second, becoming the fastest supercomputer in the world in 213.
In the list of the top 5 supercomputers in the world published on November 17th, 214, China's Tianhe II won the championship for the fourth time in a row, nearly twice as fast as the second American Titan. In May, 215, the cosmological N-body numerical simulation of neutrinos and dark matter with a population of 3 trillion was successfully carried out on Tianhe-2, revealing the long evolution process of BIGBANG since 16 million years ago. At the same time, this is the n-body numerical simulation with the largest number of particles in the world so far; On November 16, the list of the top 5 supercomputers in the world was published in the United States, and Tianhe II supercomputer was the sixth consecutive supercomputer with 3.386 trillion times per second. On June 2, 216, the new list of the top 5 supercomputers in the world was published, and the "Shenwei Taihu Light" made by China's independent chip replaced "Tianhe II" to top the list. On November 12, 218, the new list of the top 5 global supercomputers was released in Dallas, USA, and China's supercomputer "Tianhe II" ranked fourth. Basic Introduction Chinese Name: Tianhe No.2, TH-2 mbth: Tianhe-2, Milkyway-2 Attribute: Supercomputer Development Country: China Development Agency: China National University of Defense Technology Publication Date: May 213 World Ranking: Fourth (as of November 13, 218) Location: Guangzhou Supercomputing Center of Sun Yat-sen University User: Guangzhou Supercomputing Center Operating System: Kylin Linux. Composition structure, technical parameters, model, processor, front-end processor, storage, main frame, connection, operating system, energy consumption, main features, main performance, application field, research history, prospect, honor, ranking first, second, third, fourth, fifth, sixth, giving way to Shenwei, second and fourth. Each node has two processors based on Ivy Bridge-E Xeon E5 2692 and three Xeon Phis, with a total of 32, Ivy Bridge processors and 48, Xeon Phis, totaling 3.12 million computing cores. Each Xeon Phi uses 57 cores instead of all 61 cores, because using 61 cores will have problems in the coordination of operation cycles, and using 57 cores can speed up 4 threads, and the operation amount of each thread can reach 4Gflops in a single cycle. Xeon Phi running at 1.1GHz can produce double-precision operation capacity of 1.3 Tflops. If CPU is considered, 2 Ivy Bridges * .2112 TFlop/s+3 Xeon Phi * 1.3 TFlop/s = 3.431 Tflops, 2 Ivy Bridges+9 Xeon Phis, a single node can have 3.431 TFlops, and the total performance of 16, nodes can reach 54.9PFlops. Each computing node has 256GB of main memory, and each Xeon Phi has 8GB of onboard memory, so each node * * * has 88GB of memory, and a total of 16, nodes * * * have 1.44 PB of memory, while the external memory capacity is as high as 12.4PB. Tianhe II has 2 computing nodes on each motherboard, while each frame has 16 motherboards, and four frames form a cabinet. The whole system. The motherboard of each computing node is divided into two parts, one CPU and one APU. The CPU has four cores Ivy Bridge, memory and one Xeon Phi coprocessor, while the APU substrate carries five Xeon Phi coprocessors. There are five horizontally inserted connection ports between the CPU and APU, which are connected by PCI-E 2. built in Ivy Bridge. Although Ivy Bridge is built with PCI Express 3. interface, Xeon Phi only supports 2., and the single channel is 1g. The front-end processors of the computing node are 496 FT-15 processors. The FT-15 processor is developed by the National University of Defense Technology for Tianhe 1, which can be said to be the biggest harvest of Tianhe 1 project. It is a 16-core Sparc V9 architecture processor with a running frequency of 1.8Ghz, a peak performance of 144 Gflops/s and a power consumption of 65W under the condition of 4nm technology, but compared with Intel's 22nm 12-core 2.2 GHz 25 W. Tianhe-2 interconnection adopts the self-developed Express-2 internal Internet, which has 13 switches, and each switch has 576 connection ports. The connecting medium is photoelectric hybrid. The specific controller is an ASIC purpose-specific integrated circuit named NRC, which adopts 9nm process, the package size is 17.16x17.16mm, and * * * has 2577 pins. The throughput of a single NRC is 2.56Tbps. On the terminal side, the network interface also adopts NIC with similar structure, but the scale is a little smaller, which is 1.76x1.76 mm, 675 pin. It is connected by PCIE 2., and the transmission rate is 6.36 GB/s. In the case of 12, nodes, the delay is also very low, only 85us. On the computing power side, Using 14336 nodes with a total of 5GB of memory to test LINPACK, the theoretical performance is 49.19Pflops, while the actual test performance is 3.65Pflops, with an efficiency of 62.3%. This efficiency is not high, and there is great potential for optimization improvement. Of course, it may also be limited by the lack of bandwidth that Xeon phi only supports PCI Express 2.. The above test used 14,336 nodes out of 16, nodes, and used 9% of the scale, which can basically represent the overall performance of Tianhe 2. The overall power consumption of performance components (processor, memory and interconnection) of Tianhe 2 is 17.6MW, while the overall computing power is 3.65PFlops, so the calculated performance per watt is 1.935Gflops. This performance/power consumption ratio can be ranked in the top five of TOP5, and its overall performance/power consumption ratio is excellent. The overall power consumption of the system is 17.6 MW, and this power consumption does not include the cooling system such as water cooling. If the overall power consumption is considered to be as high as 24MW, Guangzhou National Supercomputing Center will use the urban water supply system to build a cooling system with high cooling efficiency, which is capable of providing the cooling capacity of the 8KW system. Tianhe-2 supercomputer system consists of 17 cabinets, including 125 computer cabinets, 8 service cabinets, 13 communication cabinets and 24 storage cabinets, covering an area of 72 square meters, with a total memory capacity of 1,4 trillion bytes, a total storage capacity of 12,4 trillion bytes and a maximum operating power consumption of 17.8 MW. Tianhe-2 operates for 1 hour, which is equivalent to 1.3 billion people using calculators to calculate for a thousand years at the same time. Its total storage capacity is equivalent to storing 6 billion books with 1, words each. Compared with the American "Titan" supercomputer, which ranked first in the world before, Tianhe II has twice the computing speed and 2.5 times the computing density, and the energy efficiency ratio is equivalent. Compared with Tianhe No.1, which was previously developed by the school, the two occupy the same area. The computing performance and computing density of Tianhe No.2 have increased by more than 1 times, and the energy efficiency ratio has increased by 2 times. The power consumption for performing the same computing task is only one third of that of Tianhe No.1.. Technical parameters model Tianhe No.2 model TH-IVB-FEP processor front-end processor storage main frame connection uses Opto Electronics Hybrid Transport Technology, and uses self-made TH Express-2 backbone topology network connection. Thirteen large routers are interconnected with each computing node by photoelectric transmission medium through 576 connection ports. The controller is named NRC, and the 9 nm process is used. The data throughput of a single controller is 2.56Tbps, and the terminal network interface is connected by PCI-E 2. interface, and the data transmission rate is 6.36 GB/s.. Operating system Kirin operating system and global resource management based on Slurm (simple Linux utility for resource management). Ubuntu Linux。 Ubuntu OpenStack runs on 256 high-performance nodes and will grow to more than 6,4 nodes in the next few months. Both OpenStack and Ubuntu's compilation tool Juju will run on Tianhe-2, enabling partners and alliance organizations of the National University of Defense Technology to quickly deploy and manage high-performance cloud environment.) Energy consumption The whole machine consumes 17,88 kilowatts. After the water cooling system is installed, the power consumption will reach 24 megawatts, regardless of whether the water cooling system is installed or not. At present, the power consumption of Tianhe No.2, which is the largest in TOP5, reaches 1 million RMB per year. Based on the data released on June 17th, the performance per watt is 1.91GFLOPS, which is still not as good as Titan's 2.143GFLOPS and IBM Sequoia's 2.177GFLOPS, but much higher than Beijing's .83GFLOPS per watt and Tianhe No.1' s .668GFLOPS per watt. Main features Tianhe-2 is the second fastest supercomputer in the world today, and its comprehensive technology is at the international leading level. It has five characteristics: first, high performance, peak speed and sustained speed have set new world records; Second, low energy consumption, the energy efficiency ratio is 1.9 billion times per watt, reaching the world advanced level; Third, it is widely used, focusing on scientific engineering calculation and taking into account cloud computing; Fourth, it is easy to use, innovatively develops heterogeneous fusion architecture, and improves software compatibility and programmability; Fifth, it is cost-effective. Main performance In June 213, Tianhe-2 surpassed Titan Supercomputer (Rpeak 27,112.5TFLOPS, Rmax 17,59.TFLOPS) with peak speed of 54,92.4TFLOPS per second and sustained speed of 33,862.7 TFlops. This achievement was submitted to TOP5 on June 17th, 213. In fact, in the earlier running test, only 9% of the 16, nodes, that is, 14,336 nodes, were used, and the operation speed of LINPACK reached the performance level of 3.65PFLOPS, exceeding 74% of the previous Titan. On November 18th, 213, the international TOP5 organization announced the latest list of the top 5 supercomputers in the world, and Tianhe II topped the list nearly twice as fast as the second American Titan. Also in June 213, Tianhe No.2 ranked sixth on the Graph5 with a score of 2,61GTEPS. Graph5 is a ranking of the data-intensive computing performance of supercomputers. The first one on this list is IBM Sequoia with a score of 15,363GTEPS. The total storage capacity of Tianhe II system is equivalent to 6 billion books with 1, words each. Assuming that everyone performs an operation every second, "Tianhe II" operates for one hour, which is equivalent to 1.3 billion people counting 1 years with calculators at the same time. Application fields Tianhe-2 has been used in biomedicine, new materials, engineering design and simulation analysis, weather forecast, smart city, e-commerce, cloud computing and big data, digital media and animation design, and will be widely used in big science, big engineering, information technology and other fields, providing important support for economic and social transformation and upgrading. Tianhe-2 has gradually shown its talents in life science, material science, atmospheric science, geophysics, cosmos, economics, large-scale genome assembly, gene sequencing, pollution control and a series of big science and big projects related to the national economy and people's livelihood. In addition, the National Supercomputing Guangzhou Center actively promotes international exchanges and cooperation, and provides high-performance computing services for foreign research institutions by using Tianhe II. It is reported that the Ministry of Science and Technology has continuously increased its investment in research and development of supercomputing application software in recent years. The related R&D investment since the Twelfth Five-Year Plan has exceeded 2 million yuan. It is expected that the investment will be further increased during the Thirteenth Five-Year Plan period. According to the special project "High-performance Multi-core CPU for Large-scale Scientific Computing", in 211, China will use FT-15CPU to build its own petascale computer-"Tianhe II". With the help of Tianhe-2' s "calculation magic", the domestic C919 large passenger plane has carried out high-precision aerodynamic calculation of the external flow field; Shanghai Institute of Pharmacology, Chinese Academy of Sciences has carried out the binding affinity evaluation of 75, small molecular compounds, and completed the in vitro and in vivo activities evaluation of more than 6 kinds of drugs. Tianhe No.2 has adapted to Guangzhou e-government data management system and cloud disk storage system, and built an efficient and reliable bearing platform for smart city construction ... Research history In 29, Tianhe No.1, the first petaflop computer in China, was successfully developed, and the Galaxy Kirin operating system supported by "Nuclear High Foundation" was developed, ranking fifth in the TOP 5 list of international supercomputers and first in Asia, and making China the first in the world after the United States. In 21, Tianhe No.1, the first petaflop supercomputer developed in China, once ranked first in the list of TOP 5 supercomputers in the world, but in 211, it was surpassed by the newly developed supercomputer "Jing" in Japan. In 212, America's Titan surpassed Japan's "Beijing". "Tianhe II" was developed by 28 people over two years, costing about 1 million US dollars. The R&D cost about 1 million US dollars, which was jointly funded by the Ministry of Science and Technology, the people of Guangdong Province and the people of Guangzhou. In the second half of 213, it will be in Guangzhou Supercomputing.