Crap! How to Get Terrible Performance from your HPC codes
Friday, May 4, 2018 - 12:30pm

What: Computational & Data Science Seminar


When: 12:30 pm May 4th, 2018


Where: Cramer Hall 418

Title: Crap! How to Get Terrible Performance from your HPC codes
We have available today a range of increasingly complex platforms designed for scientific computing, with the promise of extreme scale performance.  BUT unless we write our code with performance in mind, it is still possible for them to run poorly, no matter how fabulous the underlying platform may be.  In this talk I will provide a set of examples that demonstrate getting it wrong, explain why they perform poorly, and how to fix them.  I will show what commonly used performance tools for HPC do or do not show us about the codes,  and provide a glimpse of promising performance tool research.  
Karen Karavanic is an Associate Professor of Computer Science at Portland State University.  Her research interests center around environment-aware runtime measurement and diagnosis of large scale systems, including all levels of the software stack, the underlying hardware platform, and the surrounding data center environment.  Dr. Karavanic received her Ph.D. from the University of Wisconsin - Madison, where she was a member of the Paradyn Parallel Performance Tool Research group, a WARF Fellow, and a NASA GSRP Fellow.  She has worked at Lawrence Livermore National Laboratory, UC San Diego Supercomputing Center, and New Mexico Consortium.
Following the talk, at 1:30, in the same room:
There will be a hands-on tools demo based on the examples discussed in the talk. The demo will show OpenACC, a high level annotation used to parallelize code, and will demo the free “community version” of OpenACC. 
For more info about the overall seminar series: