Optimizing Performance on Many-Core Processors: Unleashing the Power of the Intel® Xeon Phi and Beyond
Authors: Dr. Richard Gerber (National Energy Research Scientific Computing Center)
Abstract: The first HPC systems based on the next-generation many-core Intel Xeon Phi processor (KNL) are becoming available to the community this fall. This BOF, conducted by the Intel Xeon Phi Users Group (IXPUG), will provide a forum for application and tool developers, HPC center staff, and industry experts to discuss their early successes and challenges. This BOF will showcase code optimization successes, highlighting tuning methodologies, tools features, and real-world impact and community codes. IXPUG is an independent users group for anyone interested in application performance on the Intel Xeon Phi. See http://ixpug.org
Long Description: Performance programming for many-core architectures is becoming paramount as more and larger systems become available to the HPC community. This year Intel released the next generation of its many-core processor, the Intel Xeon Phi (KNL), and new large systems are becoming available to application developers. Large supercomputers based on KNL have come online, or are in the process of doing so, at Argonne National Laboratory (Theta), Los Alamos National Laboratory (Trinity), the National Energy Research Scientific Computing Center (Cori), and the Texas Advanced Computing Center (Stampede). While the many-core architecture provides an opportunity for applications to achieve greatly improved capabilities and speed, there are challenges many application programmers must address in modernizing code to realize its benefits. This BOF will provide a forum for developers to share their early successes and challenges with many-core, with an emphasis on KNL, and engage with tool developers and performance experts from the community, HPC centers, and industry.
This BOF is the next in a line of highly successful community engagements that include BOFs at SC14, SC15, ISC 2015, ISC 2016, annual IXPUG meetings, recent workshops in Ostrava (Czech Republic) and St. Petersburg (Russia) and online working groups. IXPUG events often fill or overfill their venues, with attendance of about 150 at past SC BOFs. Within the community best practices are shared, newly developed tools and methodologies are presented, and research agendas are being set for improving performance across a wide range of application domains. Researchers are finding that others have experiences, innovations, tips and methodologies to share that they find useful in their own work – from moderately sized clusters up to the world’s largest Xeon Phi system installations.
This BOF will showcase successes in modernizing code for many-core, with an emphasis on the new Knights Landing processor, highlighting tuning methodologies, tools features, and real-world impact and community codes.
Presentations and a summary of the BOF will be made publicly available on the IXPUG web site (http://ixpug.org).
Birds of a Feather Index