Currently, there are active race supercomputers. Laboratories of different countries may seek to create more powerful supercomputers, which are intended to address some of the most important global issues including climate change, the need for car batteries with overlong period of working fusion reactor with a plasma whose temperature reaches 150 million degrees Celsius, creating bio fuel on the basis of weeds and not crops, etc.

Supercomputers allow researchers to create three-dimensional visualization, to some extent similar to modern video game that launches the endless scenario "what if" with increasingly fine detail. But even such large supercomputers, as modern, are not large enough, and one of the key issues for the 11,000 people who gathered in Portland, Ore., at 22nd annual conference on Supercomputing SC09, will be staged the following objectives: ECWA systems, that is, computers with a capacity of about ekzaflop (10 18 flop).

Productivity modern supercomputers are much less ekzaflop. According to the latest ranking of the world's fastest computers, TOP500, now in first place for performance system is Cray XT5, in which 224,256 cores. This supercomputer, which is known as Jaguar, is compiled on the basis of six-core processors AMD Opteron and is able to achieve peak performance of 2.3 petaflops (2.3h10 15 flop).

U.S. Department of Energy has already begun holding seminars on a system with a capacity of 1000 times the power of modern super computer, ECWA system. Other governments are also actively involved in building supercomputers. ECWA system will be necessary for the creation and calculation of climate models with high resolution, for the development and research of bio-energy products, design systems using fusion energy, etc. "There are serious problems ECWA-class, who simply can not be solved in a reasonable amount of time with computers that we have today", ─ spokesman said.

No matter how amazing supercomputers, they are still quite primitive, and modern designs use too much energy, space and money. Only in 1997 was first overcome the barrier of 1 teraflop, ASCI Red computer from Sandia National Laboratories has achieved performance of one trillion operations per second. In 2008, the Roadrunner computer from Los Alamos National Laboratory has achieved 1 petaflop, or a thousand trillion (one quadrillion) floating point operations per second. U.S. Department of Energy, which is responsible for funding many of the largest systems in the world, wants to be somewhere in 2011-13 were prepared, two systems, the performance of which will reach approximately 10 petaflops. However, the next major milestone in the increasing power of supercomputers is to achieve ekzaflopa, a million trillion calculations per second (one pentilon). This is 1000 times faster than a petaflop.

Probably ekzaflop will be achieved in 2018. It is expected that the great leaps of productivity will occur every ten years or so. Indirectly, this may confirm Moore's Law, under which the number of transistors on a chip will double every 18 months or so. But the problems associated with the creation of ECWA systems that go far beyond Moore's Law.

To use supercomputer Jaguar takes 7 megawatts or 7 million watts of energy. ECWA-based system would consume CPU 2GVt, or two billion watts, said Dave Turek, vice president of IBM.” This is about the size of an average nuclear power plant. This proposal is unacceptable for the future", ─ "he said. Now the problem of reducing the power consumption of supercomputers is one of the principal, together with the problem of system configuration. Researchers are increasingly inclined to use a hybrid approach combining the co-processors and accelerators with a processor in an attempt to reduce power consumption. For example, Roadrunner, which consumes 3.9 megawatts, reached slightly more than one petaflop. This supercomputer is rather complex hybrid architecture with processors AMD, Cell and synergetic processor elements.

According to experts, to achieve ekzaflopa need to create a system with the number of cores from 10 million to 100 million, more than likely the second number. However, such a system would use less memory per core, and it will require a wider memory bus. IBM today is going to create ECWA-system, using no more than 20 megawatts and accounts for 70 to 80 racks. For this purpose, the manufacturer plans to develop a hybrid system, rely heavily on the power of accelerators, which will be designed to handle specific applications. ECWA system of general purpose, no doubt, there will be, but it will happen later.