To increase the performance of processors, developers have moved from simply increasing the clock to build up the number of cores. It quickly became clear that this approach increases the performance but not linear, and there is more cores in the processor, the more difficult to achieve their agreed work. Generally is used single channel, realizing data transmission between any two cores in one direction. With a large number of core more time is spent performing computations, and waiting for some core to complete data between other pairs.
A group of researchers from the Massachusetts Institute of Technology (MIT), headed by Professor Li Shuang-Pe (Li-Shiuan Peh) proposed to implement the system within the processor interconnect arranged on the principle of the modern Internet . Cores are connected only with its immediate neighbors, and the data transfer can be done in different ways, thus avoiding congestion separate lines. Further, the shadow network interconnection is responsible for routing and chronologically ordered transmitted to the core network data packets.
Researchers have created a 36-core processor, which should demonstrate the operation of the developed system interconnects. Job prototype will be tested in such optimized cores operating system Linux. Related Products :
|