Intel第二代MIC架构处理器特性分析

Feature Analysis of the Second Generation Processor of Intel MIC Architecture

2016年6月Intel推出第二代集成众核架构的至强融核处理器(KNL)。相比第一代融核协处理器(KNC),KNL的新特性包括以下4个方面:①提供独立自启动形态,避免KNC依赖于PCI-E进行数据传输的主从结构;②提供宽度为512 bit的向量寄存器,能够一次实现16个单精度或8个双精度浮点数据的处理;③在处理器上集成高带宽内存,缓解内存访问瓶颈;④提供多种集群模式,根据程序特征对融核处理器进行划分,改善数据局部性。首先对第二代融核处理器KNL的新特性进行研究,然后利用扩展性好的典型科学计算程序对KNL的新特性进行测试分析,实验结果表明,新特性增强了KNL的并行计算能力,最后总结利用KNL进行并行优化的方法。

Intel introduced the second generation of Intel Many Integrated Core Xeon Phi processor(codenamed KNL)in June 2016.Compared to the first generation of Xeon Phi co-processor(codename KNC),KNL's new features include the following four aspects.Firstly,it provides a self-bootable pattern to avoid master-slave structure which relies on PCI-E for data transmission like KNC.Secondly,it provides 512-bit width vector registers,processing 16 single-precision or eight double-precision floating point data at one time.Thirdly,it provides high-bandwidth memory integrated in processor to break memory access bottleneck.Finally,it provides several cluster modes dividing CPU cores into several parts according to the program characteristics to improve the data locality.We analyze and study these new features of the second generation of Xeon Phi processor.Then,we carry out the experiments using typical scientific calculation program with good expansibility.The experimental results show that these new features greatly enhance KNL’s parallel computing ability.Finally we conclude the parallel optimization methods on KNL.