Beacon: A Strategic Path to Scientific Discovery Enabled by the Intel MIC Architecture

随着计算机处理器频率由于功率和密度的限制而停滞，处理器制造商正在其设计中引入额外的并行性，以获得针对高度并行应用的增加的性能。 英特尔多集成核心（英特尔MIC）架构提供了超越当前一代异构架构的潜在优势。 为了探索英特尔MIC架构在广泛的计算科学研究中的有效性，授予田纳西大学诺克斯维尔的这项合同将：（1）运行一个配备英特尔MIC协处理器的实验集群，为期两年;（2）与其他机构合作，协调相关的研究计划，旨在探索英特尔MIC架构的潜在影响，关键科学代码的编程模型。通过与英特尔的合作，该实验集群最初将配备预发布硬件，以允许各学科的领先科学家和工程师在英特尔MIC架构商业亮相之前开始使用该架构。预发布硬件将在发布时升级到商用英特尔MIC协处理器。相关的研究计划最初将包括由该项目直接支持的软件项目（来自等离子体物理学，纳米电子学，天体物理学，化学，生物学，量子色动力学和计算机科学领域），随后扩展到由全国各地的研究团队提出的更多样化的科学应用（由专家小组选择），以响应公开的参与呼吁。这些软件项目将优化重要的科学和工程应用程序，使用的技术有望在传统和基于加速器的架构上提高性能。 该项目战略性地促进了加速器硬件和并行编程模型在国家研究界的使用。相关的软件项目预计将推进最先进的技术，定位国家的研究社区，为未来的科学和工程进步。它还加深了田纳西大学诺克斯维尔与工业界的合作。最后，利用现有大学教育推广计划的相关参与活动，确保这项研究的好处在整个开放科学界得到广泛传播，并为传统上代表性不足的群体提供参与世界级计算研究的机会。

With the stagnation of computer processor frequency due to the constraints of power and density, processor manufacturers are introducing additional parallelism within their designs to attain increased performance for highly parallel applications. The Intel Many Integrated Core (Intel MIC) architecture offers potential advantages over the current generation of heterogeneous architectures. To explore the effectiveness of the Intel MIC architecture on a broad range of computational science research, this award to the University of Tennessee Knoxville will (1) operate an experimental cluster equipped with Intel MIC co-processors for a period of two years and (2) coordinate an associated research program in collaboration with other institutions aimed at exploring the potential impact of the Intel MIC architecture and related programming models on key scientific codes. Through a partnership with Intel, the experimental cluster will initially be equipped with pre-release hardware to allow leading scientists and engineers in a variety of disciplines to begin working with the Intel MIC architecture prior to its commercial debut. The pre-release hardware will be upgraded to commercially available Intel MIC co-processors upon their release. The associated research program will be comprised initially of software projects (from the fields of plasma physics, nanoelectronics, astrophysics, chemistry, biology, quantum chromodynamics, and computer science) directly supported by this project and subsequently expanded to a more diverse set of scientific applications (selected by a panel of experts) proposed by research teams across the country in response to an open call for participation. These software projects will optimize important scientific and engineering applications using techniques that are expected to benefit performance on both conventional and accelerator-based architectures. Results will be made available publicly.This project strategically promotes the use of emerging developments in accelerator hardware and parallel programming models within the nation's research community. The associated software projects are expected to advance the state of the art, positioning the nation's research community for future scientific and engineering advances. It also deepens the collaboration of the University of Tennessee Knoxville with industry. Finally, associated engagement activities that leverage existing University educational outreach programs ensure that the benefits of this research are broadly disseminated throughout the open science community and offer opportunities for traditionally underrepresented groups to participate in world-class computational research.