La Liga Serie A Bundesliga basketball NBA Yao Ming Comprehensive Chess World Championship Tennis Badminton CBA
F1 lottery < br> Special Photo
anime video game giants Milan Arsenal Barcelona Bayern's official website
blog body 108 Sharon Court, Forum Mall
microblogging sports lottery supermarket e version
Quest Milky Way One current affairs China to build world's first fast HPC TOP500 list
Page 1: Single
supercomputer is the most powerful computer, fastest, largest storage capacity for a type of computer, and more for the national high-tech fields and cutting-edge technology research, is the national scientific and technological development level and overall national strength important symbol. More than is used in large supercomputer calculation of mathematical problems, in particular kind of problem solving numerical simulation, it can help scientists to simulate climate change,mac makeup, molecular movement and even the process of nuclear testing.
we know, the super computer is the world's high-tech fields of strategic high ground, is reflected technological competitiveness and an important indicator of overall national strength. All major countries of their national scientific and technological innovation as an important infrastructure, invested heavily in research and development. However, the internal structure of a supercomputer is not a mystery, it is actually a lot of high-performance computing (HPC) forms of the client connection is made through the network, these high-performance computing clients are mostly used in recent decades in the form of the blade server form, This can effectively enhance the server compute density and lower power consumption and area units.
large-scale use of super computers
2010 年 11 released last month the world's fastest supercomputer TOP500 list, developed by the Chinese National Defense University
In fact, the development of supercomputers and a country's overall technological level are closely related, we have seen before the United States, Japan and Europe in terms of starting a supercomputer-style crazy competition, are behind this a huge computational overhead related industries driven by. China in the 80s of last century began the development effort in this area.
supercomputers in the high-performance computing (HPC) clients
1983, China's first named computer giant after the birth of the National Defense University, China has become following the United States, Japan and other countries, the ability to independently design and manufacture of supercomputers country. Then in the 90's have received a number of practical ability to supercomputers. Here let us briefly review the information in China by supercomputers advancement.
Page 2: Review of China's super-computer development
● Review of China's super-computer development
the early 90s of last century, China's high-performance computers on the market almost All imported products, China's oil core sectors of geophysical and meteorological field are even under the control of foreigners use of imported computer. 863 plans to support the Dawning series high-performance computers to explore the development and industrialization of a in the outside world, under the conditions of market economy development of China-based high-performance computers way.
1992 National Defense University developed 科技
1993 863 support in the successful development of the dawn of the computer One is China's first self-developed microprocessor chip with shared memory consisting of all symmetric multi-processor system, and create research products in China path of the server's new technology. One dawn in the implementation of multiprocessor architectures,mac brushes, operating systems and support the core of fine-grained parallel multi-threading technology in parallel to achieve a series of technological breakthroughs.
1993, China successfully developed the development of the parallel computer.
Chronicle of China's super-computer development
1995 年 company launched the dawn of the dawn of 1000, the peak speed of 2.5 billion floating-point operations per second, the actual speed of operation on the one billion floating-point operations per second, this high performance level.
1997 年 University of Defense Technology successfully developed the
1999, successfully developed the
2004 年 development, and manufacturing all parties concerned to achieve the dawn 4000A 10 trillion times per second speed of operation, this computer based on AMD Opteron processors. The main components of a 2560 CPU, 640 nodes, 5TB of memory, 95TB storage, four sets of the Internet and the dawn of cluster software.
Dawning 4000A is the first time ranked in the TOP500 list into the top ten, while AMD-based Dawning 4000A supercomputer in the Linpack efficiency is reached first in the world.
TOP500 Dawning 4000A obtained certification
2008 year ; Shuguang 5000A Dawning 5000A high performance computer using the latest quad-core AMD Barcelona (clocked at 1.9GHz) processor-based blade architecture HPP architecture, a total of 30,720 computing cores, 122.88TB memory, 700TB of data storage capacity, using low latency 20Gb Internet connectivity, the design speed of the peak floating-point operations per second, 230 trillion times, Linpack test speed forecast to reach 160T, efficiency is more than 70%.
2009 年 10 月 29 日 China's first petaflop supercomputer, Milky One is not only China's first petaflop supercomputer, and the innovative use of heterogeneous computing CPU + GPU design, not only significantly enhance the performance of theoretical calculations, and reached a very high energy efficiency. Efficiency of 431.7 MFlops / W, the current arrangement in the Green500 fifth. Milky Way One common use of multi-core Intel 6144 processor and 5120 graphics acceleration AMD processors, memory total capacity of 98TB, point to point communication bandwidth of 40Gbps, and the shared disk total capacity to 1PB.
2010 年 5 月 31 日 Dawn
2010 年 11 月 16 日 After a thorough upgrade of the The upgrade replaced the ATI GPU programmability poor products, the new concept of Fermi CUDA architecture Tesla officially entered the world's fastest supercomputers. The number of processing cores exceeded 200,000, 24,576 of last year's 8.25 times.
Page 3: GPU co-processors positioning and development
● GPU co-processors positioning and development
Although supercomputers have started to use a large area The most GPU co-processor, but the GPU from birth to be used for general-purpose processor, or experienced a long period of time. This was mainly due to the programmability of GPU development process decision.
species in the traditional GPU, Shader units from there (DirectX 8 2001 release marks the emergence Shader units) to the rapid increases in computing power (2007 Geforce 8800GTX release, general purpose computation significantly expanded the influence) after a long time. During this time, large-scale parallel computing for high-end graphics card is worthless, even a small amount of the industry's pioneers began to think and research, but also can not form the influence of the entire industry.
this stage, the super computer clusters, often have to remove , now with a lot of GPU design began to get more cheap and green computing power. CUDA's strong performance led to a general-purpose computing revolution, the revolution will dramatically change the face of the computer.
the first edition of China, 1206000000000000 times per second, China has become the world after the United States to a second petaflop supercomputer developed countries. Milky Way One common use of multi-core Intel 6144 processor and 5120 graphics acceleration processor AMD GPU, GPU which is the familiar model of the previous generation high-end GPU product ATI HD 4870X2. Deadline November 17, 2009,
far beyond the graphics rendering tasks, the use of general-purpose GPU computing research done gradually active, other than the GPU for rendering the field of computing a GPGPU (General Purpose computing on graphics processing units, GPU-based general computing.) GPGPU computing is often used CPU + GPU, heterogeneous model, but the traditional GPGPU development by way of programmability and hardware constraints, applications were limited, the development is also very difficult.
CUDA programmability advantage of excellent architecture
2007 年 6 launched the CUDA is the GPU as a data parallel computing device hardware and software system . CUDA does not need the help of graphics API, and uses a relatively easy to grasp C-language development. Developers familiar with C language from the relatively stable over from the CPU to the GPU, without having to re-learn syntax. Of course, to develop high-performance general-purpose GPU computing program, developers still need to master the parallel algorithm and a basic knowledge of GPU architecture.
NVIDIA supercomputer was able to occupy the market, and GPU into a deeper concept of general-purpose processor, in fact, not because of its superior performance, especially the theory of floating point throughput performance, but because the underlying hardware programmability of the design well able to live through the CUDA hardware and CUDA parallel computing software platform for a large number of transplant procedures. So GPU as a coprocessor to be able to rapid development in the last few years, penetration to the world famous supercomputers and high-performance computing client.
CUDA GPU strong effective use of processing power and huge memory bandwidth than the graphics rendering calculations, widely used in image processing, video communication, signal processing, artificial intelligence, pattern recognition, financial analysis, numerical calculations, oil exploration, astronomical calculations, fluid mechanics, biological computing, molecular dynamics calculations, database management, coding encryption and other fields, and on the CPU in these areas received one to two orders of magnitude faster. Has made remarkable achievements.
Page 4: Tesla and FireStream calculation unit
● Tesla and FireStream calculation unit
graphics processor (GPU) concept was proposed a decade ago, Then use a desktop computer has been very common. Powerful vector for GPU computing power and high memory bandwidth, which are the three-dimensional graphics applications for high performance is essential. But in 2001, NIVIDA company first introduced the GeForce 3 programmable vertex shader (Vertex Shader) unit marks the GPU has occurred functions. Subsequent years, scientists try to transplant the GPU numerous common software programmability of GPU while also developing rapidly.
NVIDIA and ATI in the general computing competition
NVIDIA Tesla (Chinese name: Tesla) is the NVIDIA Quadro and entertainment following the Professional Graphics Accelerator GeForce series of cards after the launch of a new product line, mainly used in the majority of high-performance computing needs of scientific research. NVIDIA Tesla altar at a computer workstation, a small computer cluster computing capacity to support the efficient use of energy into the parallel computing power. Tesla 700 patents is a famous scientist, the founder of the AC and radio, transformer, and the inventor of AC motors, low-loss high-voltage transmission put forward the concept.
GeForce desktop and high performance computing Tesla
with the GeForce and Quadro series relative, Tesla although the use of the same chip architecture, but each chip stars electrical properties are very favorable, while chips and more specialized databases are NVIDIA activated double precision capability to provide product performance is different from the desktop. Of course, Tesla product design and supply components PCB materials is strictly limited,vibram 5 fingers, so the Tesla has a lower MTBF and mean time to repair failures.
applied to the configuration is a six-core Xeon Intel 14336 X5670 2.93GHz CPU and the 7168 Nvidia Tesla M2050 GPU and 2048 self-developed eight nuclear FT FT-1000 CPU. The number of processing cores exceeded 200,000, 24,576 of last year's 8.25 times. In fact, the Milky Way One is not just upgrade, but almost updated the entire structure is used in the previous generation of Intel quad-core CPU + ATI GPU, through the Infiniband interconnect, and is Intel's next-generation six-core CPU + NVIDIA GPU + self-eight nuclear FT CPU, through the proprietary network interconnection.
Firestream AMD's brand is one of the series. And Radeon (for consumer graphics cards) and FirePro (for professional cards) different, FireStream is mainly used for high performance computing AMD card series. FireStream products are not used for 3D GPU acceleration purposes, but to use GPU's stream processors built into a group of parallel processors as a floating point co-processor to assist the central processor floating point computation complexity of the procedure such complex scientific computing. Firestream competitors is the Tesla series of high performance computing nVIDIA cards.
FireStream stream processor development software, called Stream SDK. In August 2008, AMD announced it will upgrade the software to support DirectX 11 and OpenCL. Catalyst 8.12 from the beginning, the mainstream graphics cards will be able to Stream technology, in March 2010 released AMD Stream SDK 2.1, formal support OnenCL.
FireStream accelerate
since the advent of the R520 series GPU, based on its programmable architecture, ATI put a lot of sources of GPGPU,mac makeup wholesale, which means using GPU computing for non-3D, treatment is usually in the mainstream server and desktop software running on the CPU, said to be 10-30 times higher performance than the CPU, and was later announced that its Stream Computing / General Purpose Computing) concept, also released ATI FireStream stream processor, use the name Stream processor architecture features, with the most current principles used by the processor to optimize the program.
2009 年 10 月 29 日 appearance of the first edition of With GPU reached a peak of the world's super-operator, so that China also have their own set of highly programmable high-performance GPU computing cluster. I believe that with the gradual increase GPU programmability, the future we will see more use of the super computer system unit as calculated Tesla and FireStream coprocessor, there is no doubt the future of supercomputers will have a higher density and lower operation unit power consumption.
popular sports Fantasy Open Cup champion to win the new Kobe Bryant and equipment
