PDS: Design Studies on Petaflops Special-Purpose Hardware for Astrophysical Particle Simulations

PDS：天体物理粒子模拟千万亿次专用硬件的设计研究

• 批准号: 9612029
• 负责人: Stephen McMillan
• 金额: $ 9.6万
• 依托单位: Drexel University
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 1996
• 资助国家: 美国
• 起止时间: 1996-09-01 至 1998-08-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9612029&HistoricalAwards=false
• 关键词: PDS; Design; Studies; Petaflops; Special

This project will study the feasibility of a petaflops-class computer for use in particle simulations in astrophysics and other areas of computational science. The planned system will operate by performing the most CPU-intensive part of a simulation (the computations of long-range interparticle forces) in hardware, leaving the remainder of the calculation to a front-end host computer of more traditional design. Confidence in the basic structure of the hardware is strengthened by the success of the 1.1 Tflops "GRAPE-4" system, which has a similar design philosophy and became operational on 1995. Utilizing straightforward improvements in fabrication technology, a petaflops machine is a realistic goal by the year 2000. Significant questions must be studied to refine design ideas, explore configuration option, and ensure the widest applicability of the new machine. A number of target problem areas in astrophysics, with a wide range of spatial and temporal scales, have been identified. Preliminary estimate suggest that substantial gains will be realized over the conventional general-purpose supercomputers likely to be available in the year 2000, with a time advantage of over a decade in some cases. Outside astrophysics, the subject most likely to benefit from a special-purpose computational engine of this sort are molecular dynamics and vortex fluid methods.

该项目将研究用于天体物理学和其他计算科学领域的粒子模拟的千万亿次级计算机的可行性。计划中的系统将通过在硬件上执行模拟中cpu最密集的部分（远程粒子间力的计算）来运行，而将其余的计算留给更传统设计的前端主机。1.1 Tflops“GRAPE-4”系统的成功加强了对硬件基本结构的信心，该系统具有类似的设计理念并于1995年开始运行。利用制造技术的直接改进，到2000年，千万亿次浮点运算的机器是一个现实的目标。必须研究一些重要的问题，以完善设计思想，探索配置选项，并确保新机器的最广泛适用性。已经确定了天体物理学中具有广泛空间和时间尺度的一些目标问题领域。初步估计表明，与可能在2000年问世的传统通用超级计算机相比，将实现实质性的进步，在某些情况下，时间优势将超过十年。在天体物理学之外，最有可能受益于这种专用计算引擎的学科是分子动力学和涡旋流体方法。