Intel "Knights Corner" impact Super Machine speed

Intel has launched a new chip that can run at 1 trillion times per second, the speed with which the U.S. government spent 55 million dollars to buy a supercomputer 15 years ago. One teraflop is calculated 1 trillion times per second.

The chip, known as Knights Corner, was first unveiled at the SC11 Super Computer conference in Seattle. Intel does not describe all specifications for this processor, including energy consumption or the actual number of cores (over 50). But the processor has contracted a major client, but the delivery date has been postponed to next year.

Intel Corporation Technical Computing general Manager Rajeeb Hazara describes the processor and identifies "this is a real processor."

The processor's performance is also interesting because it can run Linux workloads from a "safe location."

Intel does not specify when the Knights corner processor will be put on the market, although the processor already has at least one major customer-the Texas High Computing Center. The Austin City Center will begin installing the system next year, which is expected to be fully completed by 2013. The start phase runs at a speed of 10000 trillion times.

One way to understand Knights corner processor performance is through time.

In 1997, the Chadille National Laboratory's ASCI Red broke through the trillions of times per second, using almost 10,000 Pentium processors to achieve 1 trillion calculations per second. The total research and development cost is 55 million dollars.

In 2008, the first time the Cuckoo supercomputer at the Ormos National Laboratory first broke through petajoules. That is, the continuous floating-point computation capacity is more than 1000 trillion times per second.

If they want to achieve billions of dollars in the next 10 years, computer manufacturers must find new ways to provide computer performance at low energy consumption. Exascale is 1000 times times as petajoules times.

The new Intel processor is based on the Intel Corporation's MIC (multi-integration core) architecture, similar to a graphics processor, and is a 64-bit coprocessor designed to handle high parallel applications.

"Multi-integrated core processors should be easier to program because they are using the same instruction set architecture as the Intel X86 processor," said Steve Conway, an analyst at IDC research firm.

"With high performance computing entering an era of billions, more and more systems will develop the way in which X86 processors and accelerators are combined," Conway said.

Although Intel has previously launched a chip that broke through 1 trillion calculations, it has not been put into production.

Intel Corp. 's main competitor for this type of processor comes from Nvidia's graphics processor. The company is developing a way to make the ARM processor widely used in mobile phones.

Robert Harrison, director of the Oak Ridge National Academy of Computational Sciences, has been using earlier versions of the multi-integrated core processor, and says its advantage lies in programming. It uses the same software and editors as the X86 system.

"You can focus on optimization rather than starting from scratch to adapt to the new environment," says Harrison.

(Responsible editor: Lu Guang)