An increasing number of super-strength has been good news from the Chuan Keng Ho. IBM, it seems to maintain a permanent location near the world's top 500 supercomputer list, and it may be abandoned before a long-term crown.

However, the highest ranking supercomputer is not the same as a fan of Tiger Woods. The value of its real bottom-up found in the list browser. In order to make the top 500 in the world, by academic researchers in the United States and Europe and released this week, and now need to calculate the capacity of 17.1 TFLOPS, from 12.64 TFLOPS 6 months ago, this is to improve the performance - and prices. Just two years ago, the starting point on the list is 4 TFLOPS. And you can buy a set of similar powers as low as 10,000 dollars. With the democratization of personal computers, the increasing number of super-strength low-end is a big change is happening.

Capability of these devices can effectively run application software at the core of thousands of little money, it is possible to bring affordable, high-performance computing, and for more offices and R & D center, can benefit from them power. What help is that more and more low-end use of the GPU (graphics processing unit), which is lower than the cost of processors (central processing unit), which is very suitable for simulation, modeling of other high-performance use.

On the machine indicated Supercomputer Conference in Germany this week is headquartered in Boston, the British company, the use of CPU and GPU processor (AMD's FireStream 9250), which will be the first 1U rack for sale, but as a desktop tower. It also reached TFLOPS. Price not yet clears.

What is most interesting is the increasing activities of the super low-end computers, particularly in the language and structure to support the parallel environment, including the graphics processor chip. OpenCL, the initial formation of the Apple, is published by the end of last year of the Khronos Group consortium. And Nvidia Corp.s CUDA technology is released in 2007.

However, the use of a limiter of the GPU / CPU is not the hardware of the system, it is "the people of the progress made so far in a variety of applications ported to the GPU said," Steve Conway, IDC analysis of a high-performance division.

Conway said he believes this will continue to be sufficient to increase a few years ago to expand the adoption of the application. He predicted that within five years will reach the legal age applications, but in the near future will be more limited in the GPU generic processor. 500 in the world is still of interest to many reasons. This is a map, is one of the emerging and diversified economy. Saudi Arabia, for example, ranked 14 in the list of 65,500 from the core system IBM.

There are many ranking systems do not register. U.S. National Security Agency, to run some large systems, and finally appear in the Global 500 in 1998. However, when the races to break the petaflop record, last year to more than 1000 trillion yen (1 trillion) sustained floating-point operations per second, which brought about a new supercomputer to the territory, exaflop, this is a calculated times or quntillion, per second 1000 times faster than a petaflop.

Petaflop is the moment of explosion of high-performance computing, a similar, exaflop the moment, many years, so focus on it may mean missing the real change leader. The real revolution in access to high-performance system has just begun.