FPGA-Based Computational Accelerators - R21

描述（由申请人提供）：对更具成本效益，灵活和方便的高性能计算的需求是生物信息学和计算生物学不同领域的共同主线。我们建议通过使用低成本、基于fpga的计算协处理器来增强pc来解决这个问题；也就是说，在成本和易用性上与显卡相似的插件板。提议的硬件由主流商品组件组成。要开发的关键技术是软件环境，它将完全隐藏底层硬件，不仅对用户，而且对大多数应用程序开发人员。对于各种各样的应用程序，预计交付的产品速度比PC提高了100到1000倍，同时还为普通程序员创建新应用程序提供了一个环境。将提议的系统与现有的替代方案进行比较：个人电脑不是很强大；小集群既不强大，也不容易使用；大型集群和超级计算机非常昂贵且难以使用；而且专用硬件价格昂贵，不灵活，适用性有限。基本的创新是使用fpga来加速广泛的用户可定义应用程序。技术创新基于两个因素：电路的现场可编程性和计算族内特性的共性。这使系统能够基于模板和可调用库，使用户无需硬件知识即可访问高效的硬件设计。预期的广泛影响是桌面可用计算能力的质的提高。从长远来看，PC集群可能会出现类似的增长。特定的应用领域不仅包括序列处理，还包括微阵列数据分析、分子相互作用建模、遗传网络建模等。

DESCRIPTION (provided by applicant): The need for more cost-effective, flexible, and convenient high-performance computing is a common thread across diverse areas of Bioinfomatics and Computational Biology. We propose to address this problem by augmenting PCs with low-cost, FPGA-based, computational coprocessors; that is, with plug-in boards similar in cost and ease of use to graphics cards. The proposed hardware consists of mainstream commodity components. The key technology to be developed is the software environment, which will completely hide the underlying hardware not only from users, but also from most application developers. The projected deliverable is a factor of 100 to 1000 speed-up over a PC for a wide variety of applications, together with an environment for the creation of new applications by ordinary programmers. To compare the proposed system with current alternatives: PCs are not very powerful; small clusters are neither very powerful, nor easy to use; large clusters and supercomputers are very expensive and hard to use; and specialized hardware is expensive, inflexible, and has limited applicability. The basic innovation is the use of FPGAs to accelerate broad-based user-definable applications. The technical innovation is based on two factors: the field programmability of the circuits and the commonality of characteristics within families of computations. These enable a system based on templates and callable libraries, giving the user access to efficient hardware designs without requiring hardware knowledge. The expected broad impact is a qualitative increase in computing capability available at the desktop. The longer term may see a similar increase with respect to PC clusters. Specific areas of applicability are not only sequence processing, but also microarray data analysis, modeling molecular interactions, modeling genetic networks, and many others.