Intel’s New Chip Is Powering the World’s Number One Supercomputer
Intel’s (NASDAQ:INTC) new Xeon Phi line of high-speed computer chips are being used in the world’s fastest and most powerful supercomputer, China’s Xianhe-2.
Xianhe-2, which translates as MilkWay-2, was just named the world’s most powerful computer on a list published twice a year ranking the world’s top 500 supercomputers. The computer contains 48,000 of the Xeon Phi chips in addition to 32,000 Intel Xeon chips that are more commonly used in regular computers. Xianhe-2 offers more than double the performance of the top-rated supercomputer on the November 2012 list.
The difference between the Xeon and the Xeon Phi chips is in the number of processor cores, which handle complicated computing tasks. The original Xeon chip has between one and eight of the processors, while the new model has twelve. The Xeon and Phi chips work together to break up calculations into small pieces, passing information from the Xeon to the Phi chips to complete tasks faster.
That technique of using helper chips has been a popular way to boost productivity without creating a huge electricity bill in the process. Intel’s competitors Nvidia (NASDAQ:NVDA) and Advanced Micro Devices (NYSE:AMD) also use a similar approach, but with graphics chips, or GPUs, instead of accelerators like Intel. Intel believes this gives them an advantage, as accelerators are more similar to processors used in personal computers and so are easier to program than graphics chips.
The supercomputer market is expected to grow in annual revenue from $11 billion to $15 billion in the next four years. Supercomputers are used for a variety of tasks, including mapping the human genome, developing cures for diseases, increasing the accuracy of weather predictions, and analyzing more energy efficient resources.
Intel processors power over 80 percent of the supercomputers on the top 500 list, and the company has revealed plans for the next generation of Xeon Phi chips, which Intel believes will extend its dominance in supercomputing.