CINXE.COM

<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd"> <html xmlns="http://www.w3.org/1999/xhtml"> <head> <meta http-equiv="Content-Type" content="text/html; charset=utf-8" /> <title>ComSIS | Computer Science and Information Systems</title> <link rel="stylesheet" type="text/css" href="res/style1.css" /> </head> <body> <script type="text/javascript" src="res/wz_tooltip.js"></script> <script type="text/javascript" src="res/slide.js"></script> <div id="all"> <div id="header"> <h1>Computer Science and Information Systems</h1> </div>  <div id="main"> <div id="sidebar"> <p>About the journal</p> <ul> <li><a href="index.php">Home page</a></li> <li><a href="contact.php">Contact information</a></li> <li><a href="aims.php">Aims and scope</a></li> <li><a href="indexing.php">Indexing information</a></li> <li><a href="policies.php">Editorial policies</a></li> <li><a href="consortium.php">ComSIS consortium</a></li> <li><a href="boards.php">Journal boards</a></li> <li><a href="managing.php">Managing board</a></li> </ul> <p>For authors</p> <ul> <li><a href="information.php">Information for contributors</a></li> <li><a href="http://ojs.pmf.uns.ac.rs/index.php/comsis">Paper submission</a></li> <li><a href="submission.php">Article submission through OJS</a></li> <li><a href="copyright.php">Copyright transfer form</a></li> <li><a href="download.php">Download section</a></li> </ul> <p>For readers</p> <ul> <li><a href="archive.php?show=lstnew">Forthcoming articles</a></li> <li><a href="archive.php?show=vol2104">Current issue</a></li> <li><a href="archive.php">Archive</a></li> </ul> <p>For reviewers</p> <ul> <li><a href="http://ojs.pmf.uns.ac.rs/index.php/comsis">View and review submissions</a></li> </ul> <p>News</p> <ul> <li><a href="https://www.facebook.com/ComSISJournal/"> <img src="res/fb.png" alt="FB"/> Journal's Facebook page</a></li> <li><a href="cfp.php">Calls for special issues</a></li> <li><a href="notification.php">New issue notification</a></li> </ul> </div>  <div id="content">  <h1 class="title">Demystifying Power and Performance Variations in GPU Systems through Microarchitectural Analysis</h1><p class="authors">Burak Topcu<sup>1</sup>, Deniz Karabacak<sup>2</sup> and I艧谋l 脰z<sup>3</sup></p><ol><li>The Pennsylvania State University<br/>Department of Computer Science and Engineering, State College, PA, USA<br/>topcuuburak@gmail.com</li><li>Izmir Institute of Technology<br/>Electrical and Electronics Engineering Department, Izmir, Turkey<br/>denizkarabacak@std.iyte.edu.tr</li><li>Izmir Institute of Technology<br/>Computer Engineering Department, Izmir, Turkey<br/>isiloz@iyte.edu.tr</li></ol><h3>Abstract</h3><p>Graphics Processing Units (GPUs) serve efficient parallel execution for general-purpose computations at high-performance computing and embedded systems. While performance concerns guide the main optimization efforts, power issues become significant for energy-efficient and sustainable GPU executions. Profilers and simulators report statistics about the target execution; however, they either present only performance metrics in a coarse kernel function level or lack visualization support that can enable microarchitectural performance analysis or performance-power consumption comparison. Evaluating runtime performance and power consumption dynamically across GPU components enables a comprehensive tradeoff analysis for GPU architects and software developers. In this work, we present a novel memory performance and power monitoring tool for GPU programs, GPPRMon, which performs a systematic metric collection and provides useful visualization views to guide power and performance analysis for target executions. Our simulation-based framework dynamically gathers SM and memory-related microarchitectural metrics by monitoring individual instructions, and reports achieved performance and power consumption information at runtime. Our visualization interface presents spatial and temporal views of the execution. While the first demonstrates the performance and power metrics across GPU memory components, including global memory, cache, and SMs, the latter shows the corresponding information at the instruction granularity in a timeline. Based on our framework, we demonstrate performance and power analysis for memory-bound graph applications and resource-critical embedded programs from GPU benchmark suites. Our case studies reveal potential usages of our tool in memory-bound kernel identification, performance bottleneck analysis of a memory-intensive workload, performancepower evaluation of an embedded application, and the impact of input size on the memory structures of an embedded system.</p><h3>Key words</h3><p>GPU Computing, Performance monitoring, Power consumption</p><h3>Full text</h3><p><a class="download" href="pdf.php?id=17539-35824-2-rv"><img class="left" src="res/pdf.png" alt="Download"/>Available in PDF<br/><em>Portable Document Format</em></a></p><h3>How to cite</h3><p>Topcu, B., Karabacak, D., 脰z, I.: Demystifying Power and Performance Variations in GPU Systems through Microarchitectural Analysis. Computer Science and Information Systems</p>  </div>  </div>  <div id="footer_top"> </div> <div id="footer"> <div class="left">Faculty of Sciences, Trg Dositeja Obradovića 3, 21000 Novi Sad, Serbia, <a href="mailto:comsis@uns.ac.rs">comsis@uns.ac.rs</a></div> <div class="left">Published by ComSIS Consortium under<br/><a rel="license" href="http://creativecommons.org/licenses/by-nc-nd/4.0/">Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License<br><img alt="Creative Commons License" style="border-width:0" src="https://i.creativecommons.org/l/by-nc-nd/4.0/88x31.png"/></a></div> <div class="clearer"> </div> </div>  </div>  </body> </html>