Research Resources: A High-Performance Shared-Purpose Cluster for Computer Architectural Simulation and Perceptual Interactive Ray Tracing

EIA 02-24434Luebke，David P.Skadron，Kevin；Stan，Mircea R.弗吉尼亚大学cise RR：用于计算机体系结构模拟和感知交互光线跟踪的高性能共享目的集群这个建议，构建一个集群平台(具有高计算和内存要求，以及图形和显示硬件)来支持计算机图形和体系结构的研究，使研究交互渲染、计算机体系结构和温度感知计算的项目成为可能。具有双CPU、大内存(4 GB)、千兆网络和高端显卡的16节点机架安装式群集有望促进研究。具体地说，该基础设施支持以下三个研究项目：感知驱动的交互式光线跟踪、基于图形硬件的计算机体系结构热模拟和大内存体系结构模拟。计算机图形学研究基于人的感知模型研究新的场景渲染算法，需要灵活的(可编程的)图形硬件。这项工作研究了非传统的渲染技术，它利用了改变采样位置、密度和频率的能力。示例包括无帧(用像素的随机更新取代双缓冲)、可中断(统一空间和时间误差以达到保真度/性能折衷原则)和视线定向(根据用户的注意力改变细节和分辨率)渲染。为了研究微处理器，特别是多线程(SMT)和单片(SCC)系统的运行时热行为，热建模是为了求解热传递方程。计算机体系结构部分包括SMT和SCC系统的性能以及使用远距离相关性的积极分支预测技术的调查。教育方面包括将相关主题归入当前的研究生研讨会和研究中的学生。

EIA 02-24434Luebke, David P.Skadron, Kevin; Stan, Mircea R.University of Virginia CISE RR: A High-Performance Shared-Purpose Cluster for Computer Architectural Simulation and Perceptual Interactive Ray Tracing This proposal, building a cluster platform (with high computing and memory requirements, and graphics and display hardware) to support research in computer graphics and architecture, enables projects studying interactive rendering, computer architecture, and temperature-aware computing. A 16-node rack-mounted cluster with dual CPUs, large memory (4 GB), gigabit networking, and high-end graphics is expected to facilitate the research. Specifically, the infrastructure supports the following three research projects:Perceptually Driven Interactive Ray Tracing,Thermal Simulation of Computer Architecture with Graphics Hardware, andLarge-Memory Architectural Simulations.The computer graphics research investigates new scene rendering algorithms motivated by models of human perception, and requiring flexible (programmable) graphics hardware. This work investigates unconventional rendering techniques that exploit the ability to vary sample location, density, and frequency. Examples include frameless (replaces double buffering with randomized update of pixels), interruptible (unifies spatial and temporal error for a principles fidelity/performance tradeoff), and gaze-directed (varies detail and resolution according to user's attention) rendering. Thermal modeling is directed towards solving heat transfer equations for studying the run-time thermal behavior of microprocessors, especially multithreaded (SMT) and single chip (SCC) systems. The computer architecture component involves performance of SMT and SCC systems and investigation of aggressive branch prediction techniques using distant correlations.The educational aspects involve folding relevant topics into the current graduate seminars and students in the research.