Intel Corporation today introduced a new family of high-performance processors called Xeon Phi, which are specially optimized for Supercomputers.
Intel’s new chips Xeon Phi will release later this year and the first in this family of processors will be known as the Knights Korner. These chips are targeted for supercomputers and have more than 50 cores on a single chip that can significantly increase performance per watt of energy consumed, and in addition, create solutions, performance and scalability to tens of times higher than the current design.
Radzhib Hazra, Vice-President Intel Architecture Group, said that the next important milestone for supercomputers will achieve the ekzaflopa / sec mark in a performance, which is about almost 100 times more powerful than today’s supercomputers.Radzhib Hazra, said that as compared to Nvidia Tesla chips, the new Xeon Phi will offer a great opportunity to parallelize computations and for the complex calculations for scientific purposes. At Intel, said that the next generation Xeon Phi will become part of a wider family of solutions to be applied in the production servers to manage a wide range of tasks: cloud platforms, databases, virtualization, computer modeling, and others.
Among other things, the new chips will be used several technologies, which today are used only for ultra-fast supercomputers to exchange data between the dozens or even hundreds of nodes. chipmaker said that first Xeon Phi will not be a full-fledged CPU, and something like a coprocessor whereas at each work site should be set at least one more “normal” Xeon. First Xeon Phi will be used in a supercomputer Stampede, which created by Dell for the University of Texas. Stampede in 20% of chips will be “normal” Xeon E5, while 80% will be Phi coprocessors to perform parallel tasks and calculations that require high scalability.
John Hengeveld, director of marketing for high-solutions of Intel, said that the first Xeon Phi will provide performance 40-50 gigaflops / s per watt of energy consumed, while four or five years this figure will grow multiple times – up to 10 petaflop / s. In practice, this will create a very powerful servers in a fairly compact package.
Jim McGregor, an analyst Tirias Research, said in a 2018 capacity of a supercomputer ekzaflopa / sec, in principle, really, but these models will still be very voracious in terms of energy consumption. Significant decrease in consumption can be expected around 2020. Intel, said that the first Xeon Phi will be created on 22-nanometer process, which complies with the new chips Intel Core i.