$ 100 off a super computer platform, what kind of product is how it? It really is to worship the semiconductor and motor areas of the advance of technology, so that the past high power, high computing power of supercomputers to a more affordable way to achieve, and the derived build bigger, but the cost is higher than traditional large-scale clustered super- Many of the new computer should be lower supercomputer.
The main indicator is the supercomputer computing power in the industry will come when the reference floating point computing power, that is, flops, super-fast computer because the expression, usually to much GFLOPS, many PFLOPS units. The current market trend is the maximum performance per watt of the number is the key, that is to say, firms try to improve the handling computing power per watt, it will not waste electricity.
The super high-efficiency low-cost computer platforms
Before very red Parallella, known as a pioneer in this field, the original is Andreas Olofsson The master, in 2010 to start a company to use their own pensions developed. This super computer platforms, using multi-core processors Epiphany, Epiphany is how come?
Epiphany is actually created entirely by Andreas Olofsson, this is in his own basement under the house to do research, to put this new ultra-low-power architecture is completed, a person to complete the circuit and semiconductor circuit design, only to create the first generation of Epiphany System on Chip (SoC). He later went on to launch the third and fourth generation of the Epiphany.
Currently the market is Epiphany-III with Epiphany-IV, which has a processor core SoC inside the matrix, and there are with high-speed memory, as well as single-precision floating-point engine, and it is power.
Parallella supercomputers such as credit card size (3.4 inches x 2.1 inches)
Epiphany using fourth-generation 28-nanometer manufacturing process, can support up to 64 processor array, the third generation is 65nm, can support 16 processor array. To Epiphany-IV, in under 2 watts of power, it has 100 Gflops performance, equal to possess 50Gflops/Watt performance, low-power high efficiency of this ability is stronger than many other similar products, ranging from 10 to 100 times.
If we use generic graphics processor (GPGPU) this composition processor supercomputer popular choice to compare. First with the most people are familiar with and understand more easily the general level of NVIDIA GeForce GT6XX series of view, it's floating-point lift capacity per watt only 4 to 7Gflops, high-end version is per watt 16 ~ 18Gflops, but very expensive, it's not very good example, after all, is generally commercially available product, to super computers with graphics chip performance and price is different. When a supercomputer with NVIDIA products would be such a specialized computing with Tesla graphics processor for use collaborative computing, but the Tesla chip specifications calculations, it Gflops per watt only 16-17 only, can not and cheap compared to the Epiphany.
Therefore contrast, Epiphany series chips can be said to be more worth it. The combination of super computers to create, say, do you spend 42 Taiwan Parallella series, this can produce a total of 756 processors Parallella cluster system, nor its power to 500 watts, very ground.
Epiphany using fourth-generation 28-nanometer manufacturing process, can support up to 64 processor array, the third generation is 65nm, can support 16 processor array. To Epiphany-IV, in under 2 watts of power, it has 100 Gflops performance, equal to possess 50Gflops/Watt performance, low-power high efficiency of this ability is stronger than many other similar products, ranging from 10 to 100 times.
If we use generic graphics processor (GPGPU) this composition processor supercomputer popular choice to compare. First with the most people are familiar with and understand more easily the general level of NVIDIA GeForce GT6XX series of view, it's floating-point lift capacity per watt only 4 to 7Gflops, high-end version is per watt 16 ~ 18Gflops, but very expensive, it's not very good example, after all, is generally commercially available product, to super computers with graphics chip performance and price is different. When a supercomputer with NVIDIA products would be such a specialized computing with Tesla graphics processor for use collaborative computing, but the Tesla chip specifications calculations, it Gflops per watt only 16-17 only, can not and cheap compared to the Epiphany.
Therefore contrast, Epiphany series chips can be said to be more worth it. The combination of super computers to create, say, do you spend 42 Taiwan Parallella series, this can produce a total of 756 processors Parallella cluster system, nor its power to 500 watts, very ground.
Program development, a free open-source development tools Epiphany, built-in C language compiler, debugger multi-core programs, you can also use the Eclipse IDE, as well as OpenCL SDK, OpenCL compiler, run-time libraries.
2013 summer to autumn partial shipments
Stage, Adapteva company had provided some machines on the website of the masses raising supporters priority supply brotherhood quite similar to those of the customers.However, there are some of the best products of the project to go, before the end of summer in 2013, had to be shipped to site users are raising the masses under 6,300 of orders, in the early promotion, Parallella platform and Epiphany Series processor has laid an effective foundation.
If you want to buy this super computer platform hardware, since a single price is $ 99 , the price to buy all the more there is a certain degree of adjustment, has been open to non-original site in the fund-raising sales orders of users, but this batch priority would be more behind the fall of 2013 will be shipped.
Supercomputer speed unit:
- A MFLOPS (megaFLOPS) = one million per second (= 106) times the floating-point operations
- A GFLOPS (gigaFLOPS) = one billion per second (= 109) times the floating-point operations
- A TFLOPS (teraFLOPS) = one trillion per second (= 1012) times the floating-point operations
- A PFLOPS (petaFLOPS) = sec 1 1000000000000000 (= 1015) times the floating-point operations
- A EFLOPS (exascaleFLOPS) = sec 10 billion one hundred million (= 1018) times the floating-point operations
